In the development process of intelligent vehicles,it is a necessary and important stage that manual driving and automatic driving jointly play their roles,of which one key problem is selecting an appropriate take-over time from manual driving to automatic driving when a risky situation occurs.In order to improve the driving safety,according to the data collected from a real vehicle test,driving modes are divided into safe driving,warning driving and automatic driving,based on both the driver’s report and the risky situation levels obtained by means of the K-means clustering.Then,by selecting six impact factors ( namely,the average of speed,the time to headway,the standard deviation of steering,the standard deviation of acceleration,the distance away from the lane and the driver's experience) as the feature vectors,a driving mode selection model of intelligent vehicles is constructed based on the sequential minimal optimization ( SMO) algorithm with the radial basis function ( RBF) .Moreover,the constructed model is compared with the algorithms of ID3,RBF network and SVM.The results show that the constructed model achieves an accuracy of up to 91. 7%，which is significantly superior to those of the other three algorithms.