Median absolute deviation-universal（MAD-U） threshold algorithm of wavelet transform is widely used in denoising process of partial discharge（PD） measurement.However

it is only applicable for off-line measurement of PD pulse amplitude due to distortion of the output pulse.Therefore

a noise characteristic-based wavelet adaptive threshold algorithm（NATA） is proposed in this paper to improve the MAD-U threshold algorithm.Theory and structure of NATA are discussed.MAD-U and NATA algorithms are compared in the threshold error under different signal-to-noise ratio（SNR） and the error between wavelet reconstruction signal and original signal.Results show that: MAD-U algorithm exhibits an exponential increase with SNR in the error between the wavelet reconstruction waveform of PD pulse and the original waveform of PD pulse;the error of NATA is however very small and keeps stable

without distorting the waveform of PD pulse.Moreover

NATA is independent of the decomposition scale of wavelet transform

hence it is preferable for on-line measurement and waveform detection of PD pulses.