Since EMUs has many complicated components and non-linear factors, the speed calculation is the core issue that restrictsvehicle dynamic research. To solve this problem, a new research method based on the approximate model technology was proposed.To verify the feasibility of this method, taking the independent designed high-speed EMUs CRH380A as the research object, the 11suspension parameters of the vehicle were taken as input, and the indexes of stability and stable were taken as response, by sampling in theinputs with optimal Latin hypercube design, the response values of those sample points were calculated through the dynamic model, andaccording to the response values, the three approximate models of the vehicle dynamic were established, including Kriging approximatemodel, response surface model and RBF neural network model, as well as the error of each model was calculated. The results showedthat RBF neural network model had a good performance in fitting the vehicle dynamic so that it could meet the engineering application.Eventually, the feasibility of the technique of EMUs dynamic approximate model was proved by demonstrating the practical applicationof the approximate model.