A method for rolling bearing fault diagnosis based on the probability box (p-box) and support vector machine (SVM) with particle swarm optimization (PSO) algorithm was proposed.P-boxes were obtained by using the direct p-box modeling method based on the probability and statistics analysis of fault signals' characteristics,and the p-boxes fusion was realized by using the evidence theory.P-boxes features are different under different fault conditions,so,the features of p-boxes were extracted by different methods of p-box cumulative uncertainty measurement.A feature vector set for pattern recognition was constructed,which was then brought into the SVM whose key parameters were optimized by the PSO algorithm to realize the fault diagnosis.The experimental results indicate that the method can be used to accurately diagnose the rolling bearing faults.Comparing with the traditional feature extraction methods,the validity of the method was proved.