Abstract: In order to solve the problem of the low precision of drivers' route choice estimation, a stated preference survey was conducted on the internet to collect various drivers' route choice behaviors. Based on the findings from the surveys, seventeen potential affecting factors such as city type, region, gender, age, marital status, degree, job, full-time or non-full-time employees, monthly income, crowded level of the current route, vehicle queue length of the current route, delay ratio of the current route, knowledge of an alternate route, length ratio of an alternate route, crowded level of an alternate route, anticipated travel time saving ratio and quality of dynamic travel information were identified and applied to further study. A logit model and a probit model were adopted to evaluate the significance of these factors. Gender, age, full-time or non-full-time employees, delay ratio of the current route, knowledge of an alternate route, length ratio of an alternate route, and crowded level of an alternate route were proved to be significant variables. Then a classification and regression tree (CART) improved model for estimating drivers' route choice behaviors was built based on the significant variables, where recursive partitioning and pruning algorithms were focus on improved. The verification results showed that the model' estimating precision reaches 82%, which is 6% higher than that of the existing models. The research achievements in this paper can provide technical supports for making traffic induction plans.