Because natural source super low frequency (SLF) electromagnetic detection equipment receives wideband multi-source signal, how to decompose the signal to filter out the interference signal was a key factor for the application of natural source SLF electromagnetic detection technology. In the present article, the detection equipment developed by Peking University was used to survey the coal bed methane data in the Qinshui basin, Shanxi province, and the curvelet transform method was employed to decompose those data. The analysis results indicated that the high-frequency information coming from the decomposition is the interference signals mainly generated by lightning in the atmospheric and directly received by the detection equipment, while the low frequency signal mainly contains the target information. So the reconstructed curve based on the low-frequency information was more favorable for the interpretation of the target, compared with the original spectrum curve. But the curvelet transform method could not remove the artificial frequency signal.