As the leakage current in the safety stage of contamination insulators comprises a large quantity of noises， it is difficult to obtain the valid characteristics of signals. Four wavelet threshold de-noise methods are used respectively to filter the noises from measured leakage current with different signal-to-noise ratios（SNR）.Three characteristics， namely， the waveforms， effective value and amplitude value ratio of the triple-harmonic to first-harmonic of pre-and-post de-noised leakage current are acquired， whose de-noise effects are compared and analyzed. The optimal wavelet threshold de-noise method for the leakage current in the safety stage is selected. It is shown by analysis that when SNR is greater than 1.0， the best de-noise method is the self-adaptive threshold method for the waveforms and the effective value of leakage current in the safety stage. However， the approximation effect is basally the same for another characteristic， i.e. amplitude value ratio of triple-harmonic to first-harmonic. As a general result， it is known that the self-adaptive threshold method is the best for extracting the characteristics of the leakage current in the safety stage.