Noise signals mixed in the atmospheric electric field signals when lightning happens should be estimated before analysis, extraction and use of real atmospheric electric field signals. By calculating the mean square errors between simulated signals and real atmospheric electric field signals before and after de-noising,this paper tried to analyse the de-noising effect of different wavelet function, threshold method and wavelet decomposition level. Considering ratio of lightning atmospheric electric field signal to noise signal, selecting sym5 and db5 wavelet, combined with minimaxi and rigrsure threshold method as well as 6-7 wavelet decomposition level, a quite good de-noising effect was found out. Finally, the method is applied in de-noising the atmospheric electric field waveform data of real lightning strikes. In addition, better results in mutation detection have been got.