As for the shortcomings of the existing methods of vehicle instability determination, the study on pattern recognition of vehicle running stability was carried out, and a new method of judging the vehicle lateral stability based on K means clustering algorithm was proposed. The vehicle dynamics model was established by CarSim, and the offline clustering centers and its danger level were obtained by offline clustering analysis of vehicle running state data through K means clustering algorithm. Then, the CarSim and Simulink co-simulation platform was built and Euclidean distance between data points and cluster centroids was also calculated. Vehicle running stability criterion in Simulink was designed, and the vehicle running stability online was identified. This identification method made full use of the comparison of offline data and real time data for data mining of the vehicle running data. The simulation results show that the method can accurately and real-timely quantify the vehicle's lateral stability considering various parameters, which can provide the criterion for intervention timing and degree of control system.