Abstract: According to the characteristic of the rail flash butt welding and the time-varying curve of pressure, current and displacement recorded by GAAS80/580 welding machine,ten weld quality characters which had influence on the grey-spot flaw area in the rail flash butt welded joint, were used as input data of BP network model. The particle swarm algorithm was used to optimize the weight and threshold of BP neural network model. The results showed that the characteristic parameters can well reflect the grey-spot flaw of the welding joint. In addition, the optimized BP neural network model can predict the grey-spot flaw area of welding joint accurately.