Abstract: The traditional fuzzy clustering method clustering the fault is based on the similarity between the original data. But in the fault diagnosis of rolling bearings,that way cannot extracted the deep features of the bearing data well. Especially,in the complex conditions such as coupling faults and weak faults,it is difficult to effectively distinguish the different faults features,which results in low accuracy. In order to solve that problem,we propose the AE-IFCM bearing fault diagnosis method. In this framework,the AutoEncoder（AE）network is used to extract the deep features of the bearing fault samples,and then we utilize the improved FCM for fault diagnosis. It clusters the abstract features extracted by the AE network to maximize the utilization of the sample data and reduce the risk of the model falling into a local minimum. Experiments in the Case Western Reserve university bearing fault data collection show that AE-IFCM can improve the accuracy of bearing fault diagnosis.