Abstract: According to the unpredictability and the complexity of the railway accidents happening, the suggestions using the BP neural network to predict the railway accidents during a long period were presented. And taking the USA highway-railway crossing accidents as examples, the American highway-railway crossing accident in the next three years were forecasted using the BP neural network method and the BP neural network method optimized by the Genetic Algorithm. The forcasted results obtained using the two methods were compared. It was shown that, the BP neural network method optimized by the Genetic Algorithm could be used for the railway accident long-term prediction. In 2014, from January to April and from October to December, and in the whole 2015, was the periods that the USA highway-railway crossing accidents happening frequently.