Aiming at the problems of poor real-time performance and high error rate of power system transient stability assessment, a method of transient stability assessment based on kernel principal component analysis combined with deep belief network is proposed. Firstly, a set of eigenvectors reflecting the transient stability of power system is constructed. Secondly, the feature vector set is extracted based on kernel principal component analysis, and the dimensionality of feature vector is reduced and the redundant features are filtered. The reduced eigenvectors are transmitted to the deep belief network. Finally, training analysis is carried out. The training process includes pretraining and fine tuning to optimize network parameters, and then the evaluation accuracy of deep confidence network is improved. The simulation results of New England 10-machine 39-bus system show that the method can effectively reduce the dimensionality of input data, remove redundant features, reduce the error rate and test time of transient stability assessment, as well as accurately and quickly judge the steady state of power system.