Abstract: To overcome the difficulties caused by the low density of vehicle detectors in freeways when predicting the travel time, this study has developed a travel time prediction model for freeways by using toll data. First, a real travel time dataset was established by extracting and modifying the toll data. Then the Random Forest algorithm was used to select the significant variables from the candidate variables, and the BP neural network was used to develop the travel time prediction model. The model was verified by the certain segments of Chongqing freeway, and the experiment results show that the model developed with selected nine variables could work to predict travel time effectively and accurately in various periods; the MAPE for the whole study period in two experimental segments was 7.02% and 5.76% respectively. This research could provide traffic managers a reference while making the management measures.