Abstract: Identification of moving loads on bridges may be required in the study of the fatigue or damage mechanism of bridges, and it provides reliable data for traffic planning. A novel BP neural network-based stage identification method for moving loads on bridge is proposed. To develop this method, a finite element model of a bridge and a vehicle model with two degrees of freedom and five parameters are built, and the training samples of the neural network are numerically simulated. The position, velocity and weight of vehicles on bridge are identified step by step using the stage identification technique. In the design of neural network, the training sample set is selected using the orthogonal design method, the error performance function is modified by regularization, and the initial weights are optimized by employing the genetic algorithm. The identification for moving vehicle load on a simply supported girder is numerically simulated and verified through a model test. The results show that the proposed method may identify the moving loads on bridge on-line and real-time with high accuracy and rapid convergence, and has powerful robustness and noise-immunity.