An locomotive speed sensor fault diagnosis method based on Radial Basis Function (RBF) neural network was presented. RBF neural network predictor was built by taking common faults of photoelectric speed sensor as model. Through on-line training and realtime diagnosis, it was determined whether sensor was faulted, and then diagnostic decision methods were put forward and fault date of sensor was reconstructed. The simulation results show that dynamic characteristics of sensor are accurately simulated, which can rapidly and effectively realize speed sensor fault diagnosis.