Among various types of power quality problems, voltage sag is more common and influential. The identification of voltage sag disturbance sources has great significance to improve the power quality. Different voltage sags, caused by single voltage sag disturbance sources such as short-circuit fault, starting induction motor, transformer energization and composite voltage sag disturbance sources, are analyzed. This paper proposes to analyze identification features of composite voltage sag disturbance sources based on generalized S-transform. According to the fundamental-frequency amplitude curve and sum of amplitude curve of 2nd to 5th harmonic, the feature indices of voltage sag are constructed in terms of statistics, wave morphology, entropy and energy. Then support vector machine (SVM) is employed to perform the identification of different types of voltage sag disturbance sources. The simulation results show that using feature indices of voltage sag based on generalized S-transform is better than those based on standard S-transform in identification of voltage sag disturbance sources.