Weak UHF signal generated by partial discharge in GIS is easy to cover up by various external noises. Detection of partial discharge in GIS by using UHF method

the effective noise reduction preprocessing must be adopted. This article proposes a kind of improved method of dual-tree complex wavelet transform（DTCWT）partial discharge UHF signal denoise. This method uses DTCWT to decompose noisy signals into a series of wavelet coefficients andobtains the detail components of the signal at different scales. It also uses kurtosises of time-domain signal and of envelope spectrum to pre-filter detail components sensitively which greatly improving the follow up of maximum kurtosisdeconvolution（MKD）denoise efficiency. In order to verify the validity and reliability of this new method

an experimentis conducted about GIS partial discharge UHF signal denoise. The result shows that based on the better exerting of DTCWT shift invariance

the method of this paper is available to precisely suppress the noise interference of PD signal.Further more the screening of detail signal components can improve the noise reduction effect evaluation and maintain the original PD signal characteristics.