Abstract: TEM is sensitive to geologic body of resistivity anomaly and has high construction efficiency.It is widely used to detect mine goaf.However, TEM signals are susceptible to electromagnetic noise, resulting in reduced signal-to-noise ratio, and filtering is one of the most effective methods for noise suppression.Three filtering methods are compiled in VC++.Based on the noise suppression effect and delineation of anomaly area verified by drilling, the characteristics of three filtering methods are obtained by analysis.Pumping smooth filtering can effectively suppress the random noise with small amplitude, which is beneficial to the geological interpretation of the layered stratum, but will relatively weaken local anomalous response; slope smooth filtering can effectively suppress random noise, better suppress power frequency noise, and improve lateral resolution, but it depends on the selection of reference points; mixed wavelet filtering can effectively suppress random noise and power frequency noise, enhance abnormal response amplitude, and improve vertical and horizontal resolution.It is conducive to the interpretation of deep geological data, but its de-noising effect is related to the threshold function and wavelet basis selection.