Abstract.A new vibration feature extraction method based on improved local mean decomposition (LMD) is presented in this paper. Local mean decomposition is a novel adaptive time-frequency analysis method, which is widely used in rotating machinery fault diagnosis. However, traditional LMD decomposition results method is sensitive to noise. In order to eliminate influence of noise, Hermite-LMD is introduced. Firstly, the vibration signal is decomposed by Hermite-LMD method. Then, the fault frequency of gear is found through the envelope spectrum analysis of the first PF component. The effectiveness of the proposed method is verified by the simulation data and the practical gear fault diagnosis.