Aiming at the diverse types and complex judgments of electrical faults, an improved BP neural network model is proposed for fault diagnosis of CNC machine tools. By integrating the strong global search ability and fast optimization speed of particle swarm optimization(PSO) algorithm, the premature phenomenon that often occurs in neural network algorithms in diagnosis is improved to enhance the diagnostic ability of RBF algorithm. The algorithm corrects the problem of individual particle actions and the tendency of standard PSO and neural networks to fall into local minima. The experiment and simulation results based on MATLAB indicate that compared with traditional BP neural networks and fuzzy neural networks, PSO optimizes neural networks which have higher accuracy in fault identification and stronger generalization ability. The application of this scheme in fault diagnosis of CNC machine tools can effectively improve the efficiency of fault diagnosis.