To improve the disaster prevention capacity of high speed railway, and ensure the economy and rationality of rescue sites, the constrained optimal method was used to solve the location decision-making of rescue sites on long bridges. Three constraint conditions, including the rescue sites spacing, the risk grading and the rescue time, were researched and analyzed. Least rescue time that relief vehicles reach the rescue sites from the local rescue departments was taken as optimal object, the location decision-making model of rescue sites on long bridge was established, and an example was applied to validate the model. The results show that the rescue site should set at interval of 6 km to 10 km when the bridge length is longer than 10 km. The Bayesian network theory is a feasible method to evaluate the risk grading of the alternative rescue sites, and the Dijkstra algorithm is a good way to determine the optimal path between the local rescue departments and the rescue sites. The present model is simple and reliable, and it could solve the problem of locating rescue sites on long bridges well.