Aimed at the problem of shortage of capacity of the mathematical morphology filtering white noise, the feature extraction method was proposed in combining of singular value decomposition (SVD), local mean decomposition (LMD) and mathematical morphology. The SVD was used on the signal, the main ingredients in average were obtained after decomposition, and then the LMD was used. Main component was selected for sum refabricating, and then morphology difference filter was used to extract the frequency characteristics of the fault signal. Through numerical simulation test and local gear fault simulation experiment, the results show that this method can clearly extract the fault signal, and is compared with SVD and morphological filtering method, which has proved the effectiveness of the proposed method. 