As a new self- adaptive time- frequency analysis method,the local characteristic- scale decomposition（ LCD） is proposed lately and applied to mechanical fault diagnosis. The influence of noise to LCD is studied and a new fault diagnosis based on singular value decomposition（ SVD） de- noising and local characteristic- scale decomposition is proposed. Firstly,the phase space reconstruction of original vibration signal is carried out,and then,the noise is effectively eliminated by using singular value decomposition. The LCD of the de- noising signal is carried out and some intrinsic scale components（ ISC） are acquired,then the envelope spectrum analysis on the ISCs is carried out and the fault feature is extracted. Through the analysis of the simulation data and the bearing fault data,the effectiveness of this method is verified.