Abstract: In order to effectively suppress narrow-band interference and white noise in partial discharge (PD) noise signal, a denoising method for partial discharge signals based on improved variational modal decomposition (VMD) and median absolute deviation (MAD) method is proposed. An improved sliding waveform matching-mirroring algorithm was proposed to suppress the endpoint effect of VMD, and the weighted kurtosis index was constructed to determine the best decomposition mode number in VMD. We eliminated narrow-band interference and white noise according to kurtosis index of each mode after decomposition. MAD was introduced and combined with Savitzky-Golay smoothing filter to further denoise the residual white noise in each mode dominated by partial discharge pulse. Compared with traditional methods, the results of simulation and measurement show that the proposed method can preserve more effective information of partial discharge pulse while suppressing noise more effectively.