The invention discloses a GIS fault diagnosis method based on a spectrum energy analysis and a self-organizing competition algorithm, and is mainly used for diagnosing mechanical failures such as insulator looseness fault and insulation faults such as spike discharge and surface flashover for a GIS. The method comprises steps of collecting vibration signals in multiple periods, adaptively filtering the collected signals, and performing multi-level reconstruction on the filtered signals by using wavelet transform to obtain reconstructed signals in different frequency bands, calculating ratio ofenergy of the reconstructed signals in different frequency bands to total frequency band, extracting a energy ratio of frequency band with obvious change on characteristics as an input sample, training the input sample by the self-organized competition algorithm to obtain a weight center, thereby achieving fault classification and diagnosis of GIS faults. The method is easy to implement, and hasobvious results and high diagnostic accuracy.