A206 at the conflict of speed and reliability in the transient stability assessment of the power system, this paper constructs a improved artificial neural network(ANN) classifier of transient stability, improving the speed of calculation by avoiding the unnecessary whole-process simulation. Combining the performance indexes gained by short-time simulation, the classifier predicts the transient stability and terminates the simulation in advance. Different from traditional classifier, this paper defines an overlap region to avoid misclassification. Finally, the application in the New England power system with 10 machines and 39 nodes proves the effectiveness and reliability of this ANN classifier.