Do Facial Expressions Signal Specific Emotions?

Judging Emotion From the Face in Context

James M. Carroll and James A. Russell
University of British Columbia

Certain facial expressions have been theorized to be easily recognizable signals of specificemotions.
If so, these expressionsshould override situationally based expectations used by a person in attribut-
ing an emotion to another. An alternative account is offered in which the face provides information
relevant to emotion but does not signal a specific emotion. Therefore, in specified circumstances,
situational rather than facial information was predicted to determine the judged emotion. This pre-
diction was supported in 3 studies--indeed, in each of the 22 cases examined (e.g., a person in a
frightening situation but displaying a reported "facial expression of anger" was judged as afraid).
Situational information was especially influential when it suggested a nonbasic emotion (e.g., a
person in a painful situation but displayinga "facial expression of fear" was judged as in pain).

In the face-to-face encounters of everyday life, each person the woman was $5. One woman might feel surprised, another
monitors the emotional reactions of others. Is she pleased? Is he happy; a third might find the gift meager and feel disappointed,
annoyed? Are my listeners getting bored already? Ordinarily, and a fourth might construe the gift as a bribe and feel angry.
when one person observes the emotional reaction of another, a Suppose that the woman had been expecting an engagement
variety of sources of information are available. In this article, ring. It is even possible to imagine a story in which a gift of $5
we consider the case in which the observer has two such sources: is a coded message foretelling some disaster, just as a wife's re-
information about the expresser's situation and about the ex- ceipt of a telegram in World War II usually meant the death of
presser's facial expression. her husband. The objective situation (receipt of $5) can at best
The specific question we address is the relative influence of provide the observer with a guidepost, a generalized expectation
these two cues when each alone would suggest a different emo- about how most people would react, whereas the expresser's fa-
tional interpretation. We think that an answer to this question cial expression shows part of the actual reaction. When you, as
is interesting not only for its own sake but also for what it im- observer, have only incomplete information about the situation
plies in regard to the deeper issue of the precise message con- (you do not know what gift she has received, or what she had
mined in a facial expression. Let us begin our discussion with expected, or how she construed the situation), then facial infor-
the specific question of relative influence, To illustrate, suppose mation should become even more influential. In that case, see-
that you are the observer. You see a woman receive a gift, and ing the face of Figure 1 can force you to guess that the gift is
then you see her facial reaction to the gift. A photograph of her not the nice little present you first imagined but some horribly
expression is shown in Figure 1. What is the woman's emotional successful April fool's joke.
reaction? So, the face is normatively and psychologically p r e e m i n e n t - -
When the situation and the face are put into this sort of com- but only on certain matters. Consider a nonemotional example:
petition, the face has a normative preeminence. The situation is On the basis of the situation, you might fully expect a man to
a guide, not a guarantee. The situation can suggest how persons comply with a request (he is offered a choice between compli-
in general might typically react, but the facial reaction is part of ance and death), but his negative headshake answers the ques-
the very reaction to be judged. Different individuals can react tion. His headshake takes precedence, because a headshake sig-
differently to a given situation. Suppose that the gift given to nals " n o " On the other hand, imagine a society in which head-
shakes have no more meaning than toeshakes. The headshake
would be irrelevant and would not take normative precedence.
James M. Carroll and James A. Russell, Department of Psychology, To return to the topic of emotion, if the face provides no usable
Universityof British Columbia, Vancouver,British Columbia, Canada. information about emotion, then it would have no preeminence
This study was funded by a grant from the Social Sciences and Hu-
in the judgment of emotion; if the face signals specific emotions,
manities Research Council of Canada.
Correspondence concerning this article should be addressed to James then preeminence would include specific emotions. (To antici-
A. Russell, Department of Psychology,Universityof British Columbia, pate: In this article we take a position between these two ex-
Vancouver, British Columbia, Canada V6T IZ4. Electronic mail may tremes.) Our point now is that the face's normative preemi-
be sent via the Internet to [email protected]. nence is a tool to the researcher that can help delimit precisely
Journal of Personality and Social Psychology, 1996, VoL 70, No. 2, 205-218
Copyright 1996 by the American Psychological Association, Inc. 0022-3514/96/$3.00

205
206 CARROLL AND RUSSELL

wired and universally recognized. According to this widely held

view (Buck, 1984; Ekman, 1972; Fridlund, Ekman, &Oster,
1987; Izard, 1971 ), the face is part of the emotion. Information
about the face is therefore direct information about the specific
emotion. From this theory follows a prediction: When the ob-
server sees a clear, prototypical facial expression of a basic emo-
tion, then, given the natural preeminence of the face, this facial
signal should override any expectations derived from informa-
tion about the situation that the expresser is in. This prediction
is known as facial dominance. For example, in the case de-
scribed above, Figure 1 is said to show a universal signal of fear,
and therefore you are predicted to have judged the woman in
Figure 1 as frightened, even if you first expected the girl to bring
her joy.
(The only exceptions to the prediction of facial dominance
should be cases of deception. Thus, we exclude from further
consideration here situations in which the observer believes that
the expresser is deliberately using a facial expression to create a
false impression, as in the theater or in a situation in which
norms of politeness dictate that smiles be used to hide negative
reactions.)

A Prediction of Limited Situational Dominance

Our alternative account differs from Tomkins's ( 1962-1963 ) in
the nature of facial information. From our perspective, the face
does not signal specific emotions, but the observer does infer much
Figure I. What is this woman's emotional reaction? Photograph re- about the expresser from the face. We propose that the observer
produced by permission from Matsumoto and Ekman ( 1988). perceives two kinds of information from the face easily and auto-
matically. First, the observer perceives quasi-physical information.
That is, the observer can see from the eyes whether the expresser is
what information the face can provide: Whatever information weeping, winkin~ looking down or up, staring at somethir~ or
is preeminent tells us what the face clearly signals. looking way. The mouth shows whether the expresser is talking,
Most people are confident that they can read specific emo- shouting, yawning, laughing, smiling, or grimacing. We refer to
tions from faces. When experimental psychologists first exam- such information as quasi-physical to indicate its simplest literal
ined facial expressions under controlled conditions, they were meaning. Thus, as quasi-physical, the smile is recognized simply
therefore surprised to find disagreement among observers about as a smile--not whether it is a smile of joy, of embarrassment, of
the appropriate emotion to infer from the face alone (Feleky, nervousness, or a polite greeting.
1914). Confidence had not translated into consensus, and some Second, based in part on perceived quasi-physical features, the
writers (Fernberger, 1928; Landis, 1924, 1929) suggested that observer infers the expresser's feelings on the general dimensions
in everyday circumstances observers agree on a specific emotion of pleasantness and arousal. These two dimensions describe prim-
for a particular face only when the context supplies sufficient itive and universal aspects of emotional perception found in young
background information to pinpoint one emotion. children (Bullock & Russell, 1986 ) and across a range of cultural
That suggestion was overshadowed, however, by the work of and language backgrounds (Russell, 1991 ). Under various names,
Tomkins (1962-1963; Tomkins & McCarter, 1964) and those and with slightly different interpretations, these two dimensions
he inspired (Ekman, 1972; Izard, 1971 ). Tomkins initiated a are key concepts in various accounts of emotion (Bradley, 1994;
program of research that convinced most psychologists that at Cacioppo, Klein, Berntson, & Hatfield, 1993; Lan~ 1994; l.arsen
least some facial expressions are signals of specific "basic" emo- & Diener, 1992; Mandler, 1984; Tellegen, 1985). These dimen-
tions (happiness, surprise, fear, anger, disgust, contempt, and sions have also been useful in the study of facial expressions
sadness, plus or minus) and were created by evolution just for (Osgood, 1966; Russell & Bullock, 1986; Schlosberg, 1952). In
that purpose. In this article we re-examine the question of what the specific model used here, known as the circumflex, emotions
precise information the face can provide by reopening the ques- and related states fall in a roughly circular order around the pe-
tion of how observers respond to combinations of situational rimeter of the Cartesian space created when the two dimensions
and facial information. are treated as orthogonal axes. The cireumplex specifies the sim-
T h e Theoretical Issue ilarity between specific emotions.
Of course, observers also attribute specific emotions to oth-
A Prediction of Facial Dominance ers, but we propose that such judgments are slow, effortful, and
Tomkins (1962-1963) theorized that certain facial expres- less consensual. When a specific emotion is inferred, it is done
sions are signals of specific emotions that are biologically pre- so on the basis of the pleasantness, arousal, and quasi-physical
 DO FACES SIGNAL SPECIFIC EMOTIONS? 207

actions already perceived from the face--all of which are inter- physical information, pleasantness, and arousal. Therefore, an
preted in light of available information about the expresser's overall finding of facial dominance does not differentiate the
situation. To attribute a specific emotion to someone is to per- two accounts. In many studies, some situational influence and,
ceive a sufficient resemblance between that person's current on occasional trials, even situational dominance have been
state and a script for that emotion (Fehr & Russell, 1984). The found. Of course, whether this latter result was systematic or
script (or prototype) specifies the antecedents, feelings, physio- merely due to sampling fluctuations remains to be seen.
logical changes, and behavioral consequences of the emotion. The other reason for our skepticism is that the typical study
Antecedent--situational information--is therefore part of the contains technical problems that might have biased the results
actual meaning of the concept of a specific emotion. Offense, toward facial dominance. First, on each trial, the observer re-
threat, and loss are part of the meanings of anger,fear, and grief, ceives two pieces of information: a photograph of a facial ex-
respectively. pression and a written description of a situation. So, face versus
We call our position limited situational dominance. In ex- situation is confounded with visual versus written information.
plaining this phrase, we begin by noting that we accept com- Perhaps visual material is more salient or more easily grasped.
pletely the argument that facial information is naturally preem- Second, the observer typically is forced to choose one emo-
i n e n t - b u t only on matters of quasi-physical information, de- tion from a predetermined list of so-called basic emotions
gree of arousal, and, to some extent, pleasantness and (happiness, surprise, fear, anger, disgust, contempt, or sadness).
unpleasantness. You are told that it is night and that John is This forced-choice format does not allow the observer to judge
lying in bed. So, you guess that he is relaxed or even asleep. the expresser as unhappy, anxious, depressed, embarrassed,
Then you see John's face. His eyes are wide open and he is look- jealous, frustrated, in pain, puzzled, or any other of countless
ing about. This facial information takes precedence. You infer alternative interpretations. The problem here is not simply gen-
that John is not relaxed or asleep but is highly aroused. On the eralizability but that this response format precludes the re-
other hand, the face does not signal a specific emotion, and searcher from discovering that situational information leads the
therefore the face does not take precedence as to the expresser's observer to a kind of judgment not on the list. (Suppose that
specific emotion--hence the opportunity arises for situational faces convey basic emotions, whereas situations convey nonba-
dominance on the specific emotion, In the example just given, sic emotions [unhappy, anxious, depressed, etc.; Fernandez-
we predict that you infer from the facial information that John Dols, Wallbott, & Sanchez, 199 l]. If so, the situational option
is highly aroused, but not what, if any, specific emotion he is would be systematically excluded. We offer this suggestion as a
feeling. So, if you also learn that John has just received an hypothetical possibility, not as our belief.)
offensive letter, you might guess that he is feeling angry. If you Third, the design is typically within-subject. The observer en-
learn that John has just received a threatening telephone call,
counters exactly the same situation repeatedly on different trials.
you might guess that he is afraid. If you learn that he has just When the identical situation reappears, the only information that
lost his family in an accident, you might guess that he is suffer-
has changed is the face. The implicit demand to the observer is
ing grief.
to make judgments that are sensitive to the changed information.
Nakamura et al. (1990) explicitly told their observers to expect
Past Research different reactions to the same situation. Here are their instruc-
tions: "As you might expect, the reaction of the target persons
Evidence is available on what happens when an observer is pre-
[ expressers] are not exactly the same even to the same [ situation ].
sented with conflicting facial and situational cues. Reviewers have
presented this evidence as support for the theory of basic emotions In this experiment, we would like you to rate how the target per-
(Ekman, Friesen, & EUsworth, 1972, 1982). Ekman et al. ( 1972, sons felt about the [ situation ] . . . " (p. 1035). With these instruc-
1982) reviewed the work of Landis ( 1924, 1929) and other early tions, what could cooperative observers do other than vary their
researchers, criticizing all research that had suggested a large role judgments with variations in the face?
for situational information. More recent studies (Nakamura, Fourth, finally, and most important, the face-situation com-
Buck, & Kenny, 1990; Wallbott, 1988; Watson, 1972) have found binations have typically been created through a factorial pro-
the overwhelming facial dominance predicted by Tomkins's cess: All situations were paired with all faces. Many such com-
(1962-1963) theory. In the traditional research paradigm, named binations force the observer to think that crucial information is
the Goodenough-Tinker procedure after the two researchers who missing. If the woman in Figure 1 received a gift, why is she
first used this design, observers are presented with combinations so upset? (Indeed, the frown-gift combination is a not atypical
of facial and situational information. Even when the situation sug- example in these studies.) As the observer, you are forced to
gested a dear and intense specific emotion (such as grief over the imagine more situational information to explain the facial re-
death of a child), facial information was more influential in deter- action presented. In your imagination, the nice gift box might
mining the observer's judgment. No research using the no longer contain chocolates but a tarantula. Frijda (1969)
Goodenough-Tinker procedure has found results in favor of situ- noted this problem long ago, but its importance seems generally
ational dominance (Fernandez-Dols, Sierra, & Ruiz-Belda, 1993; to have been ignored. When the situational information is ob-
Nakamura et al., 1990). viously inconsistent with the facial information, the observer
Not surprisingly, we are not convinced. In the first place, must imagine more situational information to explain the visi-
Tomkins's (1962-1963) and our accounts both predict that fa- ble features perceived in the face. Our interpretation is that the
cial information is more influential in many circumstances. For observer imagines situational information to explain the quasi-
example, we predict facial preeminence on matters of quasi- physical features (shouting. staring, etc.) and degree of pleas-
208 CARROLL AND RUSSELL

antness and arousal o f the face. The methodological point is of that person displaying a reported facial expression of a "basic
that when the face and situation are inconsistent on these emotion." When judged alone, the story and face suggested
matters, the imagined situation, rather than the stated situation, different emotions. We used the best available candidates for
is what is psychologically real for observers and what they would universal facial expressions: photographs published by Ekman
use in determining their emotional judgment. and his colleagues. Our dependent measure was in a forced-
To summarize, we interpret the available evidence as incon- choice format. So, in many details, we used the method that has
clusive in differentiating the two theoretical accounts we have been used in past studies of this topic.
described. The studies carried out so far contain enough tech- We also introduced modifications that, if our alternative ac-
nical problems to render any conclusion doubtful, and in any count is viable, would be enough to produce clear situational
case, the available evidence can be interpreted in terms of the dominance. First, we did not form all possible combinations of
facial preeminence of quasi-physical features, pleasantness, and situation and facial expression. Rather, the situation described
arousal, rather than the facial preeminence of specific emo- in the story had to meet two theoretical criteria: It had, first, to
tions. The case of the upset woman of Figure 1 receiving a gift explain the quasi-physical features of the specific facial expres-
is not as informative as it might have first appeared. If our al- sion with which it was to be paired and, second, to fit the pleas-
ternative perspective is viable, then it should be possible to spec- antness and arousal conveyed by that face. For example, the
ify certain face-situation combinations in which the situational restaurant story narrated above was composed to be paired with
information reliably dominates in the choice of a specific emo- Matsumoto and Ekman's (1988) prototype facial expression
tion, despite the natural preeminence o f facial information and for "fear" (Figure 1 ). The story therefore had to suggest a per-
even when the face shows one o f the reported "facial expressions son in an extremely aroused and intensely unpleasant state star-
of a basic emotion." ing directly at someone or something. The story was designed to
suggest anger, because the circumplex predicts that fear and an-
Overview of the Present Study ger are sufficiently close in terms of pleasure and arousal. We
also used a between-subjects design and read our stories to the
Imagine that you are told the following story: participant. These latter changes were introduced simply to en-
sure that the observers actually attended to the story and were
This is a story of a woman who wanted to treat her sister to the not forced by the design of the experiment to vary their answers
most expensive, exclusive restaurant in their city. Months ahead, from one face to the next for a given story.
she made a reservation. When she and her sister arrived, they were
told by the maitre d' that their table would be ready in 45 minutes. In the first study, observers judged the combination of a
Still, an hour passed, and no table. Other groups arrived and were "fear" face with an anger story, an "anger" face with a fear story,
seated after a short wait. The woman went to the maitre d' and and a "sad" face with a disgust story. The emotions chosen were
reminded him of her reservation. He said that he'd do his best. close enough according to the circumplex (Russell, 1980) to
Ten minutes later, a local celebrity and his date arrived and were satisfy our criteria about similarity of pleasantness and arousal
immediately shown to a table. Another couple arrived and were but were still separate discrete emotions according to Tomkins's
seated immediately. The woman went to the maitre d" who said (1962-1963) theory. We purposely did not select the combina-
that all the tables were now full, and that it might be another hour tions of anger and disgust or of fear and surprise because the
before anything was available.
proposed facial expressions for these emotions are sometimes
"mistaken" for one another in Ekman's (1972) data. We would
You are next shown a picture of the woman in the story, the
have liked to include the combination of a "disgust" face with a
same photograph shown in Figure 1. Finally, you are asked
sad story, but we could think of no plausible sad story that
"What emotion is this woman feeling? happiness, surprise, fear,
would account for the protagonist wrinkling his or her nose,
anger, disgust, or sadness?"
presumably in response to an odor or taste.
The two theoretical perspectives outlined above make differ-
ent predictions about how observers respond to this question. In the second study, we sought to demonstrate that limited
According to Tomkins's (1962-1963) view, the woman in Fig- situational dominance is not restricted to labels for basic emo-
ure 1 is displaying a universal signal of fear and, whatever the tions. (In fact, we anticipated greater situational dominance
situation described in the story, many or most people will judge when the story could suggest a nonbasic emotion and observers
her afraid. According to our alternative view, the woman's face were allowed to choose a nonbasic emotion.) We therefore cre-
in Figure 1 is extremely aroused, she is unhappy, and she is star- ated stories that satisfied our criteria stated above but when
ing. The story explains these aspects of her facial behavior and judged alone suggested a nonbasic emotion. We used stories for
suggests that she is very angry. Therefore, many or most people puzzlement, determination, hope, and pain. The facial expres-
will judge her angry. sions remained the hypothesized signals for basic emotions--
We report three studies in which observers were given just thus Tomkins's ( 1962-1963) prediction remained clear. We hy-
such face-situation combinations that would allow clear, pothesized that, for example, the "'facial expression of anger;'
differential predictions from the two perspectives outlined when paired with a situation that alone suggests puzzlement,
above. We used the Goodenough-Tinker design because no would lead to the judgment that the expresser is puzzled.
study in which it has been used has thus far (see review by Na- In a third study, we replicated the results of Study 2 with sto-
kamura et al., 1990) found situational dominance, which we ries purposely designed to he more ambiguous. We hoped to
hypothesize for the cases we present. The observer was told a demonstrate that the results of Studies 1 and 2 could not be
story about a person and then shown a photograph supposedly attributed simply to source "clarity" (defined as the amount
 DO FACES SIGNAL SPECIFIC EMOTIONS? 209

of agreement achieved when that source is judged alone). We Results and Discussion
therefore hypothesized that, for example, the "facial expression
of anger" paired with a situation that vaguely suggests puzzle- The frequencies with which various emotion labels were cho-
ment would lead to a clear judgment that the expresser is sen (by the comparison group) in the face-alone condition, sit-
puzzled. uation-alone condition, and (by the experimental groups ) in the
face-plus-situation condition are shown in Table 1. The original
norms from Ekman and Friesen's (1976) study or Matsumoto
Study 1 and Ekman's (1988) study for each facial expression also are
shown. All three tasks yielded reasonably consistent responses:
We created three face-situation conditions: a "fear" face
Overall, the modal option was chosen 69% of the time in the
paired with an anger situation, an "anger" face paired with a
face-alone condition, 94% ofthe time in the story-alone condi-
fear situation, and a "sad" face paired with a disgust situation.
tion, and 70% of the time in the face-situation condition. For
For each, we studied two examples; in a between-subjects de-
all five facial expressions seen alone, and for all six stories heard
sign, there were thus six experimental groups in all. A compar-
alone, the modal label was as anticipated; with these data as
ison group rated each facial expression alone and each situation
justification, we consider the label predicted from each source
story alone.
alone as the label that would indicate the influence of that
source on judgment of their combination.
Method
Situational Dominance
Participants
To test our hypothesis of limited situational dominance, we ex-
Participant observers were 175 undergraduates of the University of amined the 150 responses from the six experimental groups to the
British Columbia (Vancouver, British Columbia, Canada), who re- various face-situation combinations. Participants chose the label
ceived partial course credit for their participation. Of these, 150 were
consistent with the situation 105 times, the label consistent with
randomly assigned to the experimental groups, 25 to a comparison
group.
the facial expression 17 times, and a label consistent with neither
28 times. The number of participants who chose the label consis-
tent with the situation was significantly greater than the number
Facial Expressions who chose all others combined, x2( 1, N = 150) = 24, p < .001.
The same pattern of results occurred within each of the three ex-
From collections published by Matsumoto and Ekman (1988) and
Ekman and Friesen (1976), we selected two of the facial expressions perimental conditions: Figure 2 shows confidence intervals for the
said to express anger,two fear, and one sadness. The photograph shown proportion of participants who chose each label for the three types
in Figure l was one ofthe "fear" expressions. of face-situation combination. Indeed, the same pattern of results
also occurred within each example within each condition.

Situations
Facial Influence
The situations were described in six short stories (presented in the
The 17 responses ( 1 I%) reported above that were consistent
Appendix). Each story was written to suggest one basic emotion and
was designed to be paired with a specific facial expression. Stories are with the facial expression were fewer than the number that
labeled by the emotion we intended the story to convey. would be expected (17%) if all responses were chosen in a ran-
dom fashion. Nevertheless, these 17 responses are consistent
with, although not facial dominance, some influence of the fa-
Procedure cial expression. To demonstrate this effect, we set aside the 105
responses that coincided with the situation. Of the remaining
Experimentalconditions. Participants were tested one at a time in a
session that lasted approximately 10 min. Each participant was pre- 45 cases, the 17 occasions in which the label consistent with the
sented one face-story combination. Each participant was read the story face was chosen was significantly more than the number pre-
by the experimenter (in a clear but nonemotional style, uniform across dicted by chance (45 responses divided by 5 response categories
stories) and then shown the stimulus photograph. After a few seconds, = 9; Z = 2.98, p < .01 ). The same conclusion can be seen an-
the experimenter asked "What emotion is the woman [ man ] feeling?" other way. The percentage of responses consistent with the situ-
The participant was then provided a response sheet with a list of six ation (70%) given the face-situation combination was signifi-
emotion words in the followingorder: happy, anger, sad, surprise,fear, cantly less than the percentage of responses consistent with the
and disgust and was asked to circle the one word that "best describes same situation in the situation-only condition (94%); x2( 1, N
how the individual is feeling:'
= 175) = 6.28, p < .05. Therefore, the addition of facial infor-
Comparison group. The procedure for the comparison group was
mation did have some influence on participants' judgments
similar to that of the experimental group except, for expediency,partic-
ipants rated all five facial expressions and all six stories. Participants about the combination.
rated the faces first, which were presented one at a time in a separate
random order for each participant with the constraint that, across par- Source Clarity
ticipants, each face be presented first the same number of times. The
stories were rated next. Each story was read aloud by the experimenter, Ekman et al. ( 1972, 1982) argued that a difference in the clarity
with the order determined in the same way.The story from Condition 3 of situational and facial information can bias the observer's judg-
was randomly chosen to be read first an extra time. ment toward whichever is clearer. We had hoped that the carefully
210 CARROLL AND RUSSELL

constructed stories and highly selected, posed photographs of fa- Study 2

cial expressions would be equally dear. This does not seem to have
happened. Source clarity has been operationalized as the percent- In Study 2 we sought to replicate the result of Study 1 with
age of participants who select the predicted label for that source other facial expressions, other situations, and, especially, other
when the source is presented alone (Ekman et al., 1972, 1982; emotions. Specifically, we consider non-"basic" emotions. We
Watson, 1972). In Study 1, when the stories were presented alone, created stories that were intended to suggest hope, determina-
94% of the responses were consistent with the predicted label. tion, puzzlement, and pain, and we augmented the options in
When the faces were presented alone, only 69% of the responses the dependent measure accordingly.
were consistent with the emotion labels proposed by Ekman and We created four types of face-situation combination (the
Friesen (1976) and Matsumoto and Ekman ( 1988 ). "surprise" face combined with the hope situation, "anger" face
Given the effort that went into the creation of these sets of pho- with determination situation, "anger" face with puzzled situa-
tographs, the difference in source clarity is itself an intriguing tion, and "'fear" face with pain situation), each with two exam-
finding. Unfortunately, it also seems to provide an alternative ex- pies, for a total of eight combinations. We also used a within-
planation for our finding of situational dominance. Table 1 shows subject design for two reasons: first, convenience, and second,
that the face and situation were about equally clear in four of the to demonstrate that a between-subjects design is not necessary
six experimental groups (90% for face; 91% for situation ), and the to achieve situational dominance. We did not, however, ever
face was much less clear in only two of the groups (44% for face; present the same facial expression or the same situation more
100% for situation). For the four combinationsthat paired sources than once to any given participant.
of approximately equal clarity, the situation-congruent label was
chosen 67 times (67%), and the face-congruent label was chosen Method
15 times ( 15% ). For the two combinations of unequal source clar-
ity, the situation-congruentlabel was chosen 38 times (76%), and Participants
the face-congruent label was chosen only twice (4%). According Participants were 75 undergraduates of the Universityof British Co-
to this comparison, source clarity may, indeed, have had some im- lumbia, who received partial course credit for their participation. Of
pact but clearly was not necessary for our finding of situational these, 50 were randomly assigned to experimental conditions, 25 to a
dominance. comparison group.

Table 1
Percentage o f Participants Who Chose Emotion Labels for Face Alone (F), Situation Alone (S),
and Face-Situation (F+S) Conditions: Study I

Story 1 Story 2

Emotion Norm F S F+S Norm F S F+S

"Fear" face with anger situation

Happy 0 0 0 0 -- 0 0 0
Angry 0 4 100 76 -- 0 84 60
Sad 4 0 0 0 -- 0 0 4
Surprise 4 48 0 20 ~ 8 0 20
Fear 88 44 0 4 76 88 0 0
Disgust 4 4 0 0 ~ 4 16 16

"Anger" face with fear situation

Happy 0 0 0 0 -- 0 0 0
Angry 76 44 . 0 4 85 88 0 32
Sad 0 0 0 0 -- 0 0 0
Surprise 3 16 0 16 -- 8 0 0
Fear 15 12 100 76 -- 0 100 56
Disgust 6 28 0 4 -- 4 0 12

"Sad" face with disgust situation

Happy -- 0 0 0 -- 0 0 0
Angry -- 0 0 0 -- 0 4 16
Sad 89 92 0 8 89 92 4 20
Surprise -- 4 4 0 -- 4 0 0
Fear -- 4 4 4 -- 4 0 0
Disgust -- 0 88 88 -- 0 92 64

Note. N~rm refers t~ ~ri~na~ data fr~m Ekman and Friesen~s ( ~976) study ~r Matsum~t~ and Ekman~s ( ~988) study. Pcr~entag~~f rcsp~ns~s t~
emotion label predicted by face alone or situation alone are in boldface. Dashes indicate that the figure is unavailable.
 DO FACES SIGNAL SPECIFIC EMOTIONS? 211

Facial Expressions 1.0

Fear face, Anger situation
Eight photographs of facial expressions (two open-mouth "anger,"
two closed-mouth "anger;' two "surprise" and 2 "fear") were selected 0,8
from those published by Ekman and Friesen (1976). None oftbese pho-
tographs had been used in Study 1. 0.6 T
Situations 0.4
The situations were described in four short stories (presented in the
Appendix). The hope story was created to be presented with a "sur-
prise" expression. The determination story was created to be presented
with an open-mouth "anger" expression. The puzzled story was created
to be presented with a closed-mouth "anger" expression. The pain story
0.2

0.0
Ang
TT2T
Fear Disg Surp Sad
was created to be presented with a "fear" expression. "Anger" expres- 1.0
sions were paired with different stories because of the quasi-physical Anger face, Fear situation
information shown. The open-mouthed "anger" expression was paired
with determination because the teeth are clenched, showing that the 0.8
expresser is exerting muscular effort. The closed-mouth "anger" expres-
sion was paired with puzzlement because the closed mouth results in 0.6
restricted breathing, suggesting that the expresser has momentarily fro-
zen action.
0.4

Procedure 0.2
Experimental conditions. With two facial expressions per situation,
we created two experimental groups, with 25 participants in each. The 0.0
groups received the same four stories but different facial expressions. Ang Fear Disg Surp Sad
The first group was read each story combined with one of the facial 1.0
expressions designated for that story; the second group was read the Sad face, Disgust situation
same four stories but combined with the second facial expressions des-
0.8
ignated for each story.
The procedure was similar to that used in Study 1 except that there
were four trials rather than one. Participants were tested one at a time,
and the session lasted approximately 15 min. The experimenter read
0.6
T
the story and then presented the photograph of the face. After a few
seconds, the experimenter asked "What emotion is the woman [ man ]
0.4
feeling?" The response sheet was a list of l0 emotion words in the fol-
lowing order: ang~ determination, disgust,fear, happy,hope,pain, sad,
surprise, and puzzled.
Comparison group. The 25 participants in the comparison group
0.2

0.0 T T
T
were each asked to rate all eight faces and all four stories. The procedure Ang Fear Oisg Sum Sad
was identical to that followed in Study 1.
Figure 2. 95% confidence intervals for the proportion of participants
who chose each emotion label for each face-situation condition in
Results and Discussion Study 1. Ang = anger; Disg = disgust; Surp = surprise.
Although Study 2 was a simple extension of Study 1, a com-
plication arose because participants' judgments of E k m a n and m a n and Friesen (1976). This difference is most likely due to
Friesen's (1976) facial expressions showed low clarity. As a con- a difference in response format in the two studies. Our format
sequence, we begin by examining participants' judgments of the provided all the response options they used, plus four additional
face alone and situation alone. We then turn to judgments of the options (determination, pain, hope, and puzzled), which together
combined cues. attracted 36% of endorsements. Especially interesting were the
results for the two closed-mouth "anger" expressions: Our par-
Face Alone ticipants' modal response was puzzled rather than anger. More
generally, for all facial expressions, at least 1 participant selected.
Responses to the facial expressions alone (F2 in Table 2 ) were the emotion label congruent with the situation to be paired with
less consistent than would be expected from Ekman and Frie- that facial expression. This result did not surprise us, because, on
sen's (1976) normative data (Norm in Table 2). Recognition our account, faces do not signal specific emotions and because
scores (percentage of participants giving the predicted response) the four additional labels were selected to be close to Ekman and
for the eight photographs ranged from 74 to 100 in Ekman and Friesen's in terms of pleasure and arousal. Moreover, recognition
Friesen's data ( M = 90.5) but from 0 to 84 in ours ( M = 48.5). scores derived from a forced-choice response format have been
For seven of the eight facial expressions, frequencies of the pre- found vulnerable to the exact list of options included (Russell,
dicted label were significantly lower than those reported by Ek- 1993).
212 CARROLL AND RUSSELL

¢,q
Because the apparent judgments of the face were quite
¢q different in the present study from those predicted by Ekman
and Friesen (1976), a distinction is made here between the
modal response to the facial expressions reported by Ekman
and Friesen (1976), which we refer to as the theoretical norm,
t z and the modal response found in the present study, which we
!im refer to as the empirical norm. On the basis o f the empirical
oli norms, the modal response was chosen on average by 64% o f
participants in the face-alone condition. On the basis o f the the-
2 oretical norms, the predicted response was chosen on average by
49% of participants in the face-alone condition. Ekman (1994)
O O ~ O OO used chance as the level against which he tested his prediction;
Z 49% is significantly greater than chance: 10% ( 100% divided by
10 response options), Z = 6.40, p < .001.
t~
~t
7E Situation Alone
~ 0
For judgments o f the situation alone, the percentages o f re-
O sponses congruent with the predicted emotion label were ade-
I Z
e.~ m quate, although lower (overall average of 79%) than that found
e~
in Study 1 (overall average of 94%). The hope story was judged
as hope by 84% of the participants, the determination story as
t'q
determination by 84%, the puzzled story aspuzzledby 64%, and
the pain story as pain by 84% of participants. In no case did any
o= °l Z
O
participant select for the situation the emotion label Ekman and
Friesen (1976) predicted for the face to be paired with that
e~ situation.
g~
¢q
¢o ¢q
Judgment of Combined Cues
• II Situational dominance. Across all eight face-situation con-
.~1 ~ I I~ I ditions, the 200 responses to the combined cues showed high

it
Z
agreement (modal response selected by 87% of participants).
Participants chose the label consistent with the situation 173
times, the label consistent with the theoretical norm for the face
9 times, the label consistent with the empirical norm for the
2
face 38 times, and a label consistent with none o f these three
possibilities 18 times. In this tally, judgments consistent with
~ l ~ l I~ I the empirical norm for the face are not mutually exclusive from
judgments consistent either with the theoretical norm for the
face or with the situation.
ua Whether the empirical or theoretical norms are used to de-
~._~ scribe the signal value of the face, there was clear evidence for
our hypothesis o f limited situational dominance. ~ The propor-

°1 O
Z
1~ ~ I l~I
tion o f participants' responses that were consistent with the sit-
uation was significantly greater than 50%, Z = 5.16, p < .001.
Indeed, for each face-situation condition, the label consistent
~.-~ with the situation was chosen significantly more often than any
b~ea~
¢a
The following omnibus test, and subsequent ones, include more
~ ~ than one observation per participant. The assumption of independence
I O ° I "*~ I I~l ~ .
z of observations is likely violated, although to an unknown degree. To
provide a conservative analysis, we used the number of participants
rather than the number of observations in calculating the error term.
¢=
A more trustworthy approach is to examine results from each face-
¢O
situation combination separately. We therefore did so by calculating
O
confidence intervals for each response category. The results are shown
Vm
pictoriallyin Figures 2, 3, and 4.
 DO FACES SIGNAL SPECIFIC EMOTIONS? 213

1.0
Surpriseface,Hopesituation of responses (.19) that were consistent with the empirical
norms for the face was significantly greater than that expected

0.6
0"8f t by chance (.10); Z = 2.12, p < .05. This last result, however,
is entirely due to the closed-mouth "anger" expressions being
0.4 judged as puzzled rather than as anger. In this condition, both
0.2 the face and situation were judged as puzzled, and therefore the
0.0 -I-.-I- T , T . T. .I T face and situation were confounded.

1.0
Angerface,Determinationsituation Source Clarity

o.8f t
0.6
0.4
The clarity of a particular facial expression was operationalized
in Study 1 as the percentage of responses to that face seen alone
that were consistent with the predicted label, which always corre-
0.2 sponded to the empirically obtained mode. However, in the pres-
T . -E" . - E . ~" T . T . "-E T ent data, Ekman and Friesen's (1976) predicted label for two of
0.0 ~
the facial expressions (the closed-mouth "anger" expressions) was

1.0
Angerface,Puzzledsituation not the most frequently chosen response. Because of this discrep-
ancy, the clarity of the facial information becomes ambiguous. If
Y clarity is defined with respect to Ekman and Friesen's (1976) pre-
0.6
0"8f diction, then the clarity of the two closed-mouth auger expressions
0.4 was 0%, and the overall clarity of the facial information was 49%.
0.2 If clarity is defined with respect to the modal response obtained
here, then the clarity of the dosed-mouth anger expressions was
0.0 -;- -r ~ -r -Z-.T T -X- ';-
moderate (60% for one expression, and 64% for the other), and
Fear face, Painsituation the overall clarity of the facial information was 64%. (For all four
1.0[ -V- situations, the modal response was the predicted label, and their
0.8 overall clarity was 79%.)
0.6 If we base our analysis on Ekman and Friesen's (1976) norms
0.4 (Table 2), the face and situation were equally clear in one of the
0.2 eight experimental groups, and the face was less clear in seven
groups. For the one condition of equal source clarity (n = 25 ),
0.0 -. E . r . T . . T . . .T . T .T -'~ T
the situation-congruent label was chosen 22 times, and the face-
congruent label was not chosen at all. For the seven conditions
"1" of unequal source clarity (n = 175), the situation-congruent
label was chosen 151 times, and the face-congruent label was
Figure 3. 95% confidence intervals for the proportion of participants chosen 9 times. If we base our analysis on our empirical norms,
who chose each emotion label for each face-situation condition in
Study 2. Ang = anger; Deter = determination; Disg = disgust; Surp = we must omit the two cases in which an "anger" face was paired
surprise; Puzz = puzzlement. with a puzzled situation, because the "anger" face was judged
as puzzled. Doing so leaves us with six conditions. One was of
equal source clarity and was described above. The other five
were of unequal source clarity (n = 125); for these the situa-
other emotion label (see Figure 3). Indeed, the same pattern tion-congruent label was chosen 107 times, and the face-con-
occurred for each example within each condition., gruent label was chosen 7 times. So, in either analysis, source
Consistent with our expectation of greater situational domi- clarity cannot account for our results, although the issue is more
nance here than in Study 1, the proportion of participants who troubling here than it was in Study 1.
chose the label consistent with the situation was greater in Study
2 (. 87 ) than in Study 1 (.70), x 2( 1, N = 200) = 5.32, p < .05. Study 3
Facial influence. The extent to which facial information
affected judgments of the combinations can be analyzed with Study 2 was troubled by problems of source clarity. There are
either the theoretical norms or the empirical norms. The num- two potential solutions: Come up with clearer faces or with
ber of participants (9) who chose the label predicted by Ekman more ambiguous situations. Clear facial signals of specific emo-
and Friesen (1976) was less than that expected by chance (200 tions have been sought by scientists for well over a century
responses divided by 10 response categories = 20). If we set ( Duchenne de Boulogne, 1862 / 1990; Darwin, 1872 / 1965; Fel-
aside the 173 responses that coincided with the situation, how- eky, 1914; Izard, 1971; Ekman, 1972), and we did not feel up
ever, only 27 cases remain. For this set, on the assumption that to the task. We therefore altered the situations of Study 2 to
the 27 cases are independent, the 9 occasions in which the label make them more ambiguous. We predicted that even with more
consistent with the face was chosen was significantly greater ambiguous situations, judgments of the combined cues still
than the number predicted by chance (27 responses divided by would conform to the situation. In short, Study 3 was a replica-
9 response categories = 3; Z = 3.65, p < .001 ). The proportion tion of Study 2, but with less clear situations.
214 CARROLL AND RUSSELL

Surprise face, Hope situation ger" expressions continued to provide interest. As in Study 2,
1.0 the mode for example Number 1 was puzzled (80%); 0% chose
0.8
0.6
-+- anger. Unlike Study 2, the mode for example Number 2 was
anger (36%); only 16% of choices were puzzled. We have no
0.4 explanation for this discrepancy, except the possibility of order
0.2 i T
effects created by our randomization for each participant.
0.0 T -1- T T T, ,-r,-r,
Situation Alone
Anger face, Determination situation
1.0 Each of our four situations was more ambiguous than its
0.8 ! counterpart in Study 2. Participants chose the predicted emo-
0.6 tion label significantlyless often in Study 3 (36%) than in Study
0.4 2 ( 79% ), x 2( 1, N = 50) = 9.44, p < .01. The predicted emotion
remained the modal response for three of the four stories. For
0.2 the pain story, however, the modal response was anger rather
0.0 T T T T T,-r -r. -E -1-
than pain.2
Anger face, Puzzled situation
1.0 Judgment of Combined Cues

0.6 f
0"8 T Frequency of response to the face alone, situation alone, and
0.4 face-plus-situation conditions are shown in Table 3. (Only fre-
quencies for those emotion labels consistent with facial or situ-
0.2 -~ i I --£- ational dominance are shown. In other words, Table 3 is a sim-
0.0 -2- . -1"- . "2- -£- . -i- .
plified version of Table 1 with only the most relevant rows
Fear face, Pain situation shown.) The eight face-situation conditions yielded a lower
1.0 amount of consensus (66%) than was found in Study 2 (86%).
0.8 Situational dominance. Participants in the face-plus-situa-
0.6 tion condition chose the label consistent with (our predicted
label for) the situation 132 times, with (Ekman and Friesen's
0.4 predicted label for) the face 19 times, and with neither 49 times.
0.2 The proportion of responses that was consistent with the situa-
0.0 tion was significantly greater than .50; Z = 2.26, p < .05. This
overall pattern was found for each face-situation condition (see
Figure 4) and, indeed, within each example as well.
Facial influence. The number of responses that were con-
Figure 4. 95% confidence intervals for the proportion of participants sistent with Ekman and Friesen's (1976) predictions did not
who chose each emotion label for each face-situation condition in differ from chance ( 19 responses evidenced, and 20 expected by
Study 3. Ang = anger; Deter = determination; Disg = disgust; Surp = chance). If we set aside the 132 responses that coincided with
surprise; Puzz = puzzlement. the situation, however, only 68 cases remain. On the assumption
that these 68 cases are independent, the 19 occasions in which
the label consistent with the face was chosen was significantly
more than the number predicted by chance (68 responses di-
Method vided by 9 response categories --- 7.5; Z = 4.46, p < .001 ).
The method of Study 3 was exactly the same as that of Study 2 except
for the situational stories (see Appendix), which had been altered so Source Clarity
that when they were presented alone, consensus as to the predicted emo-
tion would be lower. We intentionallyreduced the clarity of the situational stories.
The clarity of the faces remained similar to that of Study 2. In
Results and Discussion consequence, the faces (56%) were clearer than the situations

Face Alone
Responses to each facial expression alone are shown (as F3) 2 Nevertheless, we continue to use pain as our predicted label for the
in Table 2. Overall, 55% of participants' choices of an emotion face-situation combination. In all three studies, we used the predicted
label were consistent with Ekman and Friesen's (1976) predic- label for the situation as our predicted label for the face-plus-situation
combination as well, so that our tests were conservativeand consistent.
tion. The 55% was significantlygreater than chance: 10% ( 100%
Moreover, we believedour prediction. As we discuss later in the article,
divided by 10 response options, Z = 7.50, p < .001). Although we do not believe that each source, face and situation, is judged sepa-
slightly greater than the comparable figure found in Study 2 rately as to emotion. It is the combined information that is judged. In
(49%), it was not significantly so and remained well below the this case, the high arousal and unpleasantness of the face in the situation
figures reported by Ekman and Friesen. The closed-mouth "an- described in the story are most plausibly interpreted as pain.
 DO FACES SIGNAL SPECIFIC EMOTIONS? 215

Table 3
Percentage of Participants Who Chose Emotion Labels in Face-Only (F), Situation-Only (S),
and Face-Situation (F+S) Conditions, Study 3
Example 1 Example 2
Emotion label Norm F S F+S Norm F S F+S

"Surprise" face with hope situation

Surprise 96 80 4 20 74 72 4 20
Hope -- 4 56 68 -- 0 56 56

"Anger" face with determination situation

Anger 100 64 0 0 100 80 0 8
Determination -- 16 40 84 -- 12 40 68
"Anger" face with puzzled situation
Anger 74 0 4 0 92 36 4 16
Puzzled -- 80 44 84 -- 16 44 56

"Fear" face with pain situation

Fear 100 52 0 8 88 60 0 4
Pain -- 0 4 48 -- 16 4 64
Note. The percentage of responses is given only for the emotion labels predicted by Ekman and Friesen's
(1976) normative data or by our predicted interpretation of the situations. Norm refers to original data from
Ekman and Friesen (1976). Dashes indicate that data were not collected.

(36%). Indeed, the situation was less clear than the face in six interesting cases occur when face and situation are consistent in
of the eight conditions. (The remaining two cases involved the anticipated quasi-physical features, pleasure, and arousal, but dis-
troublesome closed-mouth "anger" expression.) Nevertheless, crepant on specific emotions.
despite the superior clarity of the face, all eight combinations Unlike previous studies, therefore, we did not pair all faces with
showed situational dominance. all situations. Rather; for each facial expression we created situa-
tions that matched it in anticipated pleasantness, arousal, and
General Discussion quasi-physical information--but of course differed in anticipated
specific emotion. Within this limited context, most observers
Like everyone else, we believe that a person often looks to an-
judged the expresser to be feeling the emotion anticipated from the
other's face when trying to discover that person's emotional reac-
situation rather than the one allegedly signaled by the face. This
tion. But what precise information do observers derive from the
basic result recurred in 22 of the 22 face-situation combinations
face? Attempts to bring facial expressions under scientific scrutiny
examined in three studies.
have created controversy around this question, including the tra-
Our finding of limited situational dominance is difficult to
ditional controversy of categories versus dimensions. As students
reconcile with the theory that the specific facial expressions in-
in introductory psychology know, one important theory of emo-
cluded in these 22 cases signal specific emotions (Ekman, 1972;
tion teaches that certain facial expressions are easily recognized
signals of specific emotions. Although common sense, this theory Izard, 1971; Tomldns, 1962-1963). If faces signal specific emo-
has not always been accepted (landis, 1924, 1929; Woodworth, tions, and if facial information is normatively preeminent over
1938), and it is once again becoming controversial (Fridlund, situational information, then our empirical results are baffling.
1994; Ortony & Turner, 1990; Russell, 1994; but see replies by Another, serendipitous, finding that is hard to reconcile with the
Ekman, 1992, 1994; and Izard, 1992, 1994). same theory was the change in modal response to those same
Part of the controversy surrounds the appropriate methods to facial expressions when the response format was changed from
be used in the study of faces (Ekman, 1994; Izard, 1994; Russell, 6 to 10 options. The 4 items that we added--hope, determina-
1994). In this article, we resurrected one traditional method used tion, inpain, andpuzzled---drew a fair number of endorsements
to approach the problem: We examined judgments of faces paired in competition with the supposedly "correct" answer.
with situations. We relied on the face's normative preeminence Two findings in our studies might appear to lend at least some
over situations to help delimit the precise information the observer support to the hypothesis that faces signal specific emotions. First,
derives from the face. On our account, the face's normative pre- observers often selected the emotion label predicted by Ekman
eminence is limited to quasi-physical features, pleasure, and and his colleagues for the face when seen alone. These high "rec-
arousal. So, both our account and those ofTomkins ( 1962-1963), ognition scores" for faces seen alone replicate a ubiquitous result
Izard (1994), and Ekman (1994) predict facial dominance when (Ekman & Friesen, 1976; Matsumoto & Ekman, 1988). The in-
face and situation are discrepant with regard to these matters. The terpretation rather than the reliability of this result is the question.
216 CARROLL AND RUSSELL

We interpret high recognition scores as partly due to the pleasant- when one or the other result occurs: When face and situation
ness, arousal, and quasi-physical properties of the face (e.g,, the are incongruent on quasi-physical information, pleasure, or
nose wrinkle of the "disgust expression" conveys an unpleasant arousal, then facial information will take its natural precedence
reaction to an odor; not all emotions terms are equally applicable over situational information. When face and situation are con-
to this reaction) and partly due to the method of gathering judg- gruent on quasi-physical information, pleasure, and arousal,
ments. That is, the forced-choiee response format, within-subject then situational information will determine the precise emotion
design, and other features of method used here and in previous category chosen.
studies help funnel the observers'judgment of the face into a tingle It is now time to turn to other designs. We especially need
category (for a discussion of these issues, see Russell, 1994). As we research designs with more ecological relevance. The kind of
saw, the recognition scores vary with small changes in the response still photographs of posed facial expressions paired with ver-
format. bally described situations does not occur outside psychologists'
There is a second finding that might seem to suggest that faces experiments. We very much need information on the kinds of
signal specific emotions: When judging the face-situation com- facial movements that actually occur in daily life, about the
binations, a very small but still significant number of observers kinds of situations in which they occur, and on the ways in
in each study chose the emotion label predicted as the signal which observers spontaneously interpret such information.
value of the face. We believe that the combination of forced- Laboratory experiments obviously have their place in answer-
choice response format and the quasi-physical features of the ing theoretical questions about what can happen, but these ex-
face account for this effect. The forced-choice response format periments must be supplemented with information about what
limited the available options severely. Tightened lips, clenched does happen in the nonexperimental world of face-to-face
teeth, staring eyes, and the like can suggest the behavioral as- encounters.
pects of the expresser's emotional reaction. For example, Frid- In the future, new questions as well as methods need to be
lund(1994) hypothesized that such facial actions are social sig- explored. For example, our studies did not address the question
nals: Stares are felt as threats, tears are felt as requests for help, of what process intervenes between the presentation of the stim-
and smiles are felt as greetings. In a complementary approach, ulus information and the observer's judgment (see Wallbott,
Frijda (1986) hypothesized that facial actions are incipient in- 1988). Our studies do provide some suggestions, however. Pre-
strumental actions, such as approach, withdrawal, biting, and vious writing on this topic has generally assumed that the ob-
so on. Observers who focus on such quasi-physical features server judges each source separately as to the specific emotion it
might infer aspects of the situation that were not stated in o u r suggests, then weights each possible emotion (presumably by
stories and that would, in turn, suggest a somewhat different the clarity of its source), and then decides in favor of the emo-
emotional reaction. Our own account of the meaning of the face tion with the largest weight. Our data speak clearly against this
is in great need of descriptive data on how quasi-physical fea- weighted linear model of the process, which must predict that
tures of the face are interpreted. Recall as well that our observ- the judgment of the combined cues could not achieve greater
ers were college students enrolled in a psychology course. Col- consensus than either source judged alone. All three studies here
lege students in Western cultures are familiar with certain facial yielded violations of this prediction. More important, Study 3
expressions as symbols (emblems) for specific emotions: just found a clear case in which the emotion predicted from the
as a red heart stands for love, a smile stands for happiness (or combination was qualitatively different from the modal re-
sponse to each source alone. No matter what weights are as-
comedy), wide open eyes stand for surprise, and a downturned
signed to the emotions from each source alone, no such quali-
mouth stands for grief (or tragedy). These same students might
tative shift could occur.
have encountered the theory and photographs of Ekman (1972)
As an alternative to a weighted linear model, we suggest that
and Izard ( 1971 ) in their textbook.
the information from the two sources is combined before any
Ekman et al. (1972) pointed to source clarity in their criti-
specific emotion is judged. The face yields what information it
cism of previous studies that seemed to suggest a greater influ-
can, and the situation does likewise. The facial reactions are
ence from situational information than consistent with their
then explained on the basis o f situational information, and a
theory. We agree, of course, that observers may be biased toward
plausible story is created, perhaps including additional infor-
whatever source is clearer. However, Study 3 (and internal anal- mation not directly observed, but inferred. Finally, the range of
yses in Studies l and 2) showed that our prediction of limited plausible emotions compatible with the imagined story is in-
situational dominance was upheld in cases of equal source clar- ferred. If one emotion is asked for, the most plausible is chosen.
ity or even when the face was substantially dearer. Although If the experimenter forces the observer to choose from a re-
source clarity is undoubtedly a factor in this domain, it does not stricted list, then the most probable emotion from the list is
account for the influence o f the situation seen here. chosen.
Our studies deliberately relied heavily on a research design
used for decades. We did so with reluctance and, indeed, feel References
that the Goodenough-Tinker design has now reached the end
Bradley, M. M. (1994). Emotional memory: A dimensional analysis. In
of the line. It is a design at best capable of telling us what can
S. H. M. van Goozen, N. E. Van de Poll, & J. A. Sergeant (Ecls.),
happen (as opposed to what does happen ordinarily). With this Emotions: Essays on emotion theory (pp. 97-134). Hillsdale, NJ:
design, facial dominance (Nakamura et al., 1990; Watson, Edbaum.
1972 ) and situational dominance (present studies) can both be Buck, R. (1984). The communication o f emotion. New York: Guilford
demonstrated. Overall, we believe that a simple rule specifies Press.
 DO FACES SIGNAL SPECIFIC EMOTIONS? 217

Bullock, M., & Russell, J. A. (1986). Concepts of emotion in develop- Izard, C. E. (1992). Basic emotions, relations among emotions, and
mental psychology. In C. E. Izard & P. B. Read (Eds.), Measuring emotion-cognition relations. PsychologicalRevie~, 99, 561-565.
emotions in infants and children (Vol. 2, pp. 203-237 ). Cambridge, Izard, C. E. (1994). Innate and universal facial expressions: Evidence
England: Cambridge University Press. from developmental and cross-cultural research. Psychological Bulle-
Cacioppo, J. T., Klein, D. J., Berntson, G. G., & Hatfield, E. (1993). tin, 115, 288-299.
The psychophysiology of emotion. In M. Lewis & J. M. Haviland Landis, C. (1924). Studies of emotional reactions: II. General behavior
(Eds.), Handbook of emotions (pp. 119-142). New York: Guilford and facial expression. Journal of Comparative Psychology, 4, 447-
Press. 509.
Darwin, C. ( 1965 ). The expression of the emotions in man and animals. Landis, C. (1929). The interpretation of facial expression in emotion.
Chicago: University of Chicago Press. (Original work published Journal of General Psychology, 2, 59-72.
1872) Lang, P. J. (1994). The varieties of emotional experience: A meditation
Duchenne de Boulogne, G. B. (1990). The mechanism of human facial on James-Lange theory. Psychological Review, 101, 211-221.
expression. (R. A. Cutherbertson, Ed. and Trans.). Cambridge, En- Larsen, R. J., & Diener, E. (1992). Promises and problems with the
gland: Cambridge University Press. (Original work published 1862 ) circumplex model of emotions. In M. S. Clark (Ed.), Review of per-
Ekman, P. (1972). Universals and cultural differences in facial expres- sonality and social psychology ( Vol. 14, pp. 25- 59 ). Newbury Park,
sions of emotion. In J. K. Cole (Ed.), Nebraska Symposium on Mo- CA: Sage.
tivation (Vol. 19, pp. 207-283). Lincoln: University of Nebraska Mandler, G. ( 1984). Mind and body: Psychology of emotion and stress,
Press. New York: W. W. Norton.
Ekman, P. (1992). Are there basic emotions? PsychologicalReview, 99, Matsumoto, D., & Ekman, P. (1988). Japanese and Caucasian facial
550-553. expressions of emotion (JACFEE). Unpublished slide set and bro-
chure, Department of Psychology, San Francisco State University.
Ekman, P. (1994). Strong evidence for universals in facial expressions:
Nakamura, M., Buck, R., & Kenny, D. A. (1990). Relative contribu-
A reply to Russell's mistaken critique. Psychological Bulletin, 115,
tions of expressive behavior and contextual information to the judg-
268-287.
ment of the emotional state of another. Journal of Personality and
Ekman, P., & Friesen, W. V. (1976). Pictures offacial affect. Palo Alto, Social Psychology, 59, 1032-t039.
CA: Consulting Psychologists Press. Ortony, A., & Turner, T. J. (1990). What's basic about basic emotions?
Ekman, P., Friesen, W. V.; & Ellsworth, P. (1972). Emotion in the hu- Psychological Review,, 74, 315-341.
man face. New York: Pergamon. Osgood, C. E. (1966). Dimensionality of the semantic space for com-
Ekman, P., Friesen, W. V., & Elisworth, P. (1982). What are the relative munication via facial expressions. Scandinavian Journal of Psychol-
contributions of facial behavior and contextual information to the ogy, 7, 1-30.
judgment of emotion? In P. Ekman (Ed.), Emotion in the human face Russell, J. A. (1980). The circumplex model of affect. Journal of Per-
(2nd ed., pp. 111-127). New York: Cambridge University Press. sonality and Social Psychology, 39, 1161-1178.
Fehr, B., & Russell, J. A. (1984). Concept of emotion viewed from a Russell, J. A. ( 1991 ). Culture and the categorization of emotion. Psy-
prototype perspective. Journal of Experimental Psychology: General, chological Bulletin, 11O, 426-450.
113, 464-486. Russell, J. A. (1993). Forced-choice response format in the study of
Feleky, A. M. (1914). The expression of the emotions. Psychological facial expression. Motivation and Emotion, 17, 41-51.
Review, 21, 33-41. Russell, J. A. (1994). Is there universal recognition of emotion from
Fernandez-Dols, J. M., Sierra, B., & Ruiz-Belda, M. A. (1993). On the facial expression? A review of the cross-cultural studies. Psychologi-
clarity of expressive and contextual information in the recognition cal Bulletin, 115, 102-141.
of emotions: A methodological critique. European Journal of Social Russell, J. A., & Bullock, M. (1986). Fuzzy concepts and the perception
Psychology, 23, 195-202. of emotion in facial expression. Social Cognition, 4, 309-341.
Fernandez-Dols, J.-M., Wallbott, H., & Sanchez, E (1991). Emotion Schlosberg, H. (1952). The description of facial expressions in terms of
category accessibility and the decoding of emotion from facial ex- two dimensions. Journal of Experimental Psychology, 44, 229-237.
pression and context. Journal of Nonverbal Behavior, 15, 107-123. Tellegen, A. ( 1985 ). Structures of mood and personality and their rele-
Fernberger, S. W. (1928). False suggestion and the Piderit model. Amer- vance to assessing anxiety, with an emphasis on self-report. In
A. H. Tuma & J. D. Maser (Eds.), Anxiety and the anxiety disorders
ican Journal of Psychology, 40, 562-568.
(pp. 681-706 ). Hillsdale, N J: Erlbaum.
Fridlund, A. J. (1994). Human facial expression: An evolutionary view.
Tomkins, S. S. ( 1962-1963 ). Affect, imagery, consciousness (Vols. I and
San Diego, CA: Academic Press.
2). New York: Springer.
Fridlund, A. J., Ekman, P., & Oster, H. (1987). Facial expressions of Tomkins, S. S., & McCarter, R. (1964). What and where are the pri-
emotion: Review of the literature, 1970-1983. In A. W. Siegman & S. mary affects: Some evidence for a theory. Perceptual & Motor Skills,
Feldstein (Eds.), Nonverbal behavior and communication (pp. 143- 18, 119-158.
224). Hillsdale, N J: Erlbaum. Wallbott, H. G. (1988). Faces in context: The relative importance of
Frijda, N. H. (1969). Recognition of emotion. In L. Berkowitz (Ed.), facial expression and context information in determining emotion
Advances in experimental social psychology (Vol. 4, pp. 167-224). attributions. In K. R. Scherer (Ed.), Facets of emotion (pp. 139-
New York: Academic Press. 160). Hillsdale, N J: Erlbaum.
Frijda, N. H. (1986). The emotions. New York: Cambridge University Watson, S. G. (1972). Judgment of emotion from facial and contextual
Press, cue combinations. Journal of Personality and Social Psychology, 24,
Izard, C. E. ( 1971 ). Theface of emotion. New York: Appleton-Century- 334-342.
Crofts. Woodworth, R. S. ( 1938 ). Experimentalpsychology. New York: Holt.

(Appendix follows on next page)

218 C A R R O L L A N D RUSSELL

Appendix

Situation Descriptions Used in the Studies

Situations From Study 1 the bag. The bag was so full that as she carried it to the curb it tore
slightly and she could feel liquid from the bottom o f the bag drip down
Anger."Example 1. This is a story o f a m a n who had recently bought her leg.
a new car. Today, he is walking back to his car across the parking lot after
r u n n i n g errands at the post office. From a distance, he can see some kids
around his car. T h e n he sees one o f t h e m holding one o f the car's hub- Situations From Study 2
caps. He yells at the kids and they take offto a nearby forest waving the Hope. This is a story o f a w o m a n who went to the horse races to bet
hubcaps in their hands. Now that he is close to his car he can see that it her last five h u n d r e d dollars. She bet it all on horse n u m b e r 7. She is now
is certainly missing its hubcaps. watching the horses make the final turn down the stretch to the finish
Anger." Example2. This is a story o f a w o m a n who wanted to treat line. Horse n u m b e r 9 and horse n u m b e r 7 are neck and neck. It looks
her sister to the most expensive, exclusive restaurant in their city. like horse n u m b e r 7 is going to take the lead.
Months ahead, she made a reservation. W h e n she and her sister arrived Determination. This is a story about a woman who made it onto the
at the restaurant, they were told by the maitre d' that their table would Olympic rowing team. She is now in the race for the gold medal. Half-
be ready in 45 minutes. Still, an hour passed, and no table. Other groups
way through the race she is third, but gaining on second. Two thirds o f
arrived and were seated after a short wait. T h e w o m a n went to the mai- the way through the race she sees that she is in position to pass the boat
tre d' and reminded h i m o f her reservation. He said he would do his in first. She rows as fast as she ever has in her life.
best. Ten minutes later, a local celebrity a n d his date arrived and were Puzzled. This is a story o f a w o m a n who is going to a McDonalds
immediately shown a table. A n o t h e r couple arrived and were seated for the first time. While waiting in line, she stares up at the menu. Sev-
immediately. The w o m a n again went to the maitre d', who said that all eral people are served and without realizing it she arrives at the front o f
the tables were now full, and that it might be another hour before any- the line. The cashier says, "'May I help you please?" The woman ac-
thing was available. knowledges the cashier and returns her attention to the menu.
Fear: Example 1. This is a story about a m a n who is on vacation Pain. This is a story of a w o m a n who has just had her ingrown toe-
with his family. He decided to go for a hike while the rest went down to
nail operated on. She is now waiting in line to see a movie. As another
the beach for the afternoon. He enjoyed walking through the quiet movie ends, the crowd is asked to step back to let them through. Some-
shaded m o u n t a i n side. He followed a small brook in and out of large
one accidentally steps on her toe.
rocks and crevices. Without realizing he stumbled into a small cave
which the brook m u s t have been flowing through, and only about five
yards in he sees some small bear cubs. He turns and sees the adult bear Situations From Study 3
coming through the entrance o f the cave. He backs away slowly as the
Hope. This is a story o f a w o m a n who went to the horse races to bet
bear approaches h i m growling loudly. The bear has h i m cornered.
five h u n d r e d dollars. She bet it all on horse n u m b e r 7. For her to win,
Fear: Example 2. This is a story about a w o m a n who had never the horse has to finish first or second. She is now watching the horses
done anything really exciting in her life. O n e day she decided she had to
make the final turn down the stretch to the finish line. Horse n u m b e r 7
do something exciting so she enrolled in a class for parachuting. Today
is in second place.
is the day that she will make her first j u m p . She and her class are seated
Determination. This is a story about a woman who had worked very
in the plane as it reaches the right altitude for parachute j u m p i n g . T h e
hard and made it onto the Olympic rowing team. She is now in the race
instructor calls her name. It is her turn to j u m p . She refuses to leave her
for the gold medal. Halfway through the race she is in first place out of
seat. W h a t emotion is the w o m a n feeling?
five boats.
Disgust: Example 1. This is a story about a w o m a n who was earning Puzzled. This is a story o f a w o m a n who went to McDonalds for
a few dollars helping her teacher organize the biology lab. Her job was
lunch. After a short wait in line it is her turn to order. The cashier says,
to count the contents of different containers stored in boxes in the stor-
"May I help you please?" The woman acknowledges the cashier and
age room. The list of items to count ranged from frogs and worms, to
returns her attention to the menu.
h u m a n brains. T h e j o b was going quickly until she opened the container
Pain. T h i s is a story o f a w o m a n w h o went to the c i n e m a to see a
of h u m a n brains. The container was so full that she would have to take
movie. She is now waiting in line to see the movie. Just as they are
out the brains and put t h e m in a new container to get a proper count.
letting the audience into the c i n e m a , a pair o f individuals rudely p u s h
She put on a rubber glove and began to i m m e r s e her hand into the liquid
their way t h r o u g h the crowd b u m p i n g the w o m a n a n d stepping on
to pull out the first brain.
her bruised toe.
Disgust: Example 2. This is a story o f a w o m a n who went away on
quite a long business trip. W h e n she arrived home, even at the front
door, she could smell something was wrong. As she entered the kitchen R e c e i v e d F e b r u a r y 8, 1995
the smell grew even stronger. She found that she had forgotten to take R e v i s i o n r e c e i v e d J u l y 28, 1995
out the kitchen garbage. The rancid smell whooshed out as she closed A c c e p t e d A u g u s t 8, 1995 •