Abstract: A bearing fault diagnosis method that vibration signal feature extraction of AR model based on intrinsic time scale decomposition (ITD) and combined with support vector data description (SVDD) was proposed in this paper aiming to solve the problem,aiming at the problem that extracting the vibration signal characteristics of rolling bearings under strong background noise was difficult and the accuracy of fault diagnosis has affected because of the lack of fault data in the practical operation.Firstly vibration signal were decomposed into a series of proper ration (PR) components according to ITD,an AR model was then established by using each PR components,model parameters and variance of remnant were picked to construct feature vectors,which were used to set up the SVDD model of bearing normal operation,and judging the bearing running status by the degree of vibration signal feature vectors deviates from the SVDD model.The method was applied to fault diagnosis of rolling bearing and the experiment results have shown the effectiveness of the proposed method.