A new approach based on modified S-transform (MST) was proposed to detect and classify the voltage sags. The average power spectrum of white noise was deduced on the S transform and MST, and the results showed the MST'restraint upon noise. The selection thinking of adjusting parameter in MST was described, the method to get the amplitude was obtained, and the time and phases of the sag were given. Then the voltage sags in short-circuit fault , induction-motor starting and transformer exciting without load were analyzed, six feature parameters were extracted and decision rules were obtained to recognize the sag disturbance sources. The simulation results show that the amplitude and time of the voltage sags by using MST are more accurate than those by using S transform, and the classification method has a high classification ratio.