In order to monitor and identify the fault type of the automobile pump bearing on line, the internal structure and common faults of WR3258152 automobile water pump bearing as a research object are analyzed. According to the common faults, four kinds of defects of automobile water pump bearings were preset. On the established signal acquisition experimental platform, using acceleration sensor, the vibration signal of four kinds of defect bearings in operation were collected. Using MATLAB software for fast Fourier transform and frequency domain eigenvalue calculation of vibration signal, the Radial basis function (RBF) and particle swarm optimization (PSO) method were used to establish support vector machine (SVM) model, and tested and verified. The results show that the SVM classification can identify four kinds of common defects of automobile pump bearings accurately. It provides reference for on line monitoring and fault diagnosis of automobile water pump bearing.