Abstract: In fault diagnosis of rotating machinery,the strong noise and outliers interference are usually contained in the vibration signals. After fault feature extraction,the method of traditional support vector machine( SVM) for the pattern recognition causes the fuzzy of optimal hyperplane and affects the classification results. So a fuzzy C-means( FCM) clustering algorithm was introduced in this paper. FCM was used to solve the problem of fuzzy membership. However,the FCM had its own defects. The clustering result was sensitive to the initial center,and often cannot achieve the result of the global optimal. Improved by particle swarm optimization( PSO) which has advantages of global optimization search,the FCM achieved better fuzzy memberships for each sample. So,the fault diagnosis algorithm of rotating machinery based on the improved fuzzy support vector machine( FSVM) was proposed. First,fault features were extracted by using the empirical mode decomposition( EMD). Second,the problem of fuzzy membership was solved by using FCM which was optimized by PSO. At last the fuzzy memberships were put into SVM,the improved FSVM was founded and fault recognition was realized.Resultsof the experiment show that the improved FSVM has better anti-noise performance and the classification effect is better than that of the traditional FSVM algorithm.