Abstract: In this paper, a novel Immune Clone Feature Selection Algorithm (ICFSA) is proposed for fan fault diagnosis. The time wave structure features, wavelet analysis features and auditory spectrum features of real fan noise collected in the factory production line are extracted. The proposed method is compared with genetic algorithm in classification and feature selection experiments. The experimental results show that: (1) the classification accuracy of support vector machine classifier decreases a very little while the number of features is reduced 61% by the proposed method and the classification time is much shorter; (2) the proposed algorithm can converge to a more optimal feature subset faster than genetic algorithm. The results demonstrate that the proposed algorithm is an effective and robust feature selection method, and useful for fan fault diagnosis.