This paper presents a method of feature selection for transient stability assessment based on binary particle swarm optimization(BPSO). Every particle in the swarm stands for a selected subset of features. The fitness of particle is defined as the correct classification percentage by least square support vector machine (LS-SVM), which uses the selected subset of features to classify the corresponding training set. First, this paper uses the BPSO to complete the feature selection, then inputs the selected features into LS-SVM classifier for transient stability assessment. It is tested on the EPRI-36 bus model of PSASP, the result indicates that the method can evidently decrease the dimensions of input features while greatly increase the correct classification percentage.