Four typical defects are made in GIS

the UHF partial discharge signals of these defects are acquired

and a denoising method for these UHF PD signals is proposed. The empirical mode decomposition（EMD） is used to decompose the UHF PD signals of the four defects into a finite number of the intrinsic mode function（IMF） components

the cut-off point between the PD signal and the white noise inthe IMF component is determined by using the normalized autocorrelation function

and the IMF component with white noise is filtered by using the wavelet soft threshold. Then

all the IMF componentsare reconstructed to obtain the partial discharge signals. Compared with the wavelet denoising method

theproposed method obtains the UHF PD signals with higher signal-to-noise ratio and slighter waveform distortion.