Abstract: Accident rate of curve roads in mountain areas is often higher than which in general sections.Due to sight distance and some reasons,drivers cannot easily identify risks of curve roads.A method for early warning of curve roads is developed.A simulation is set up based on driving modes of front vehicles and road conditions in mountain areas,in order to provide valid information and complete chain transfer of information.The relations between velocity,acceleration,and curves are analyzed.Frequency diagrams and distribution of correlation coefficients are computed.The results show that four variables taken into this model as independent variables is feasible.An early warning model is developed by using the historical data and Elman recurrent neural network.Compared with two binary logistic regression and BP neural network,the accuracy rate of this model is 85.59%.The validity of the model is verified.