In view of the actual situation that voltage sags occur frequently with diverse categories, which makes it difficult to identify the disturbance sources, a novel identification approach of voltage sag disturbance sources was proposed by combing the time-frequency characteristic of voltage sag disturbance signals, the grey wolf optimization (GWO) and support vector machine(SVM) in this paper .Multiresolution time-frequency analysis was applied to voltage sag disturbance signals by S transform, extracting the feature curves of signals from S transform result matrix, and then 8 features were calculated from 6 kinds of voltage sag complex disturbance signals. A one versus rest (OVR) GWO-SVM classifier whose inputs were fed with the extracted features was established to identify voltage sag disturbance sources. The proposed method was validated to be effective to identify the voltage sag disturbance sources by the analysis result of voltage sag simulation model based on MATLAB/Simulink, which could be also a necessary technical support for voltage sag disturbance governance.