The details in the high frequency area of power quality disturbances signals are always destroyed by the signal de-noising methods. A new approach based on S-transform and local optimal global threshold has been proposed in this paper. Firstly, the time-frequency matrixes of disturbances signals were calculated by the S-transform. Secondly, the amplitudes of S-transform modular matrix were transformed to grey level presentation. Then, the optimal global threshold was chosen by Otsu’s algorithm, and the thresholds were used to de-noise the disturbance signals. Finally, the time-frequency matrix after de-noising was transformed into disturbances signals by invert S-transform. The simulation results show that the new method can remain the high frequency elements of the disturbances with excellent de-nosing ability. It could satisfy the requirements of power quality disturbances analysis.