The insulation deterioration of mining dry-type transformer is always accompanied with partial discharge（PD）. The assessment accuracy of transformer insulation performance is directly affected by the authenticity of the PD signal. Based on the ensemble empirical mode decomposition（EEMD）and combining the principle of higher order statistics（HOS）

a hybrid method for suppressing white noise in PD signals is proposed. Firstly

the PD signals is decomposed into a number of intrinsic mode function（IMF）components. Then the signal’s Gaussian components are detected and removed by kurtosis as gaussianity estimators supplemented by Bootstrap techniques.The white noise in the components were suppressed through thresholding and reconstructing each intrinsic mode function with adaptive thresholds. Finally

the results are reconstructed and return to the original signal. The validity of this method is proved by simulated signals and practical partial discharge signals.