A power system transient stability assessment method according to comprehensive Support Vector Machine (SVM) classification model based on key sample set is presented to improve the performance of SVM classification. A construction method of comprehensive SVM classification model based on different features, a producing method of key sample set, and procedures of SVM classification based on comprehensive classification model and key sample set are given. A 3-generator 9-node system and a certain provincial power grid are analyzed to verify the effectiveness of the method. Analysis results show that the presented method largely decreases the false classification number of unstable test samples compared with the traditional SVM method, thus improves the practicability of the method. The thought of using key sample set to form the classification model might be significative to other data mining methods. This work is supported by Science and Technology Project of State Grid Corporation of China (No. 5442xt170016).