This paper presents a comparative study of feature selection methods for Arabic text classification, focusing on five methods: Ichi square, Chi square, Information gain, Mutual information, and Wrapper, tested against five classification algorithms. The research utilized a dataset consisting of 9,055 Arabic documents classified into twelve categories and evaluated the methods based on precision, recall, F-measure, and time to build the model. Results indicated that the improved Ichi square feature selection method outperformed the other methods in nearly all test cases, demonstrating its effectiveness in addressing feature selection challenges in Arabic classifiers.