The results of feature extraction of a signal depend on the selection of wavelet base when wavelet transform is employed to extract the signal features . To overcome the limitation, a method based on lifting wavelet is introduced to reconstruct the wavelet according to the characteristic of gearbox’ fault feat ure wave. The results show that the lifting wavelet can match the fault feature signal better than the general wavelet and is more effective in extracting the f ault feature information, which provides favorable condition for subsequent feat ure detection of vibration signal and fault diagnosis.