Abstract: Aiming at the complexity of 25 Hz phase sensitive track circuit fault, this paper proposed a fault diagnosis method combining simulated annealing (SA) algorithm with BP neural network.The learning process of BP neural network was optimized by using the global optimization feature of SA algorithm to avoid long training time and falling into local minimum.Through Matlab simulation analysis, the results show that the application of this method to track circuit fault diagnosis can effictively improve the efficiency and accuracy of fault diagnosis.