After the gear fault occurs， the vibration signal collected includes fault impact， deterministic meshing signal and noise signal， and these signals are also affected by the transmission path， which makes the gear fault feature extraction difficult. Cepstrum analysis is a common method for gear fault diagnosis. It can display the periodic components in the sideband as a single line， which is helpful for fault diagnosis. However， when the fault characteristic signal is weak， the fault characteristics in the cepstrum are not obvious. For this， feature-enhanced cepstrum analysis method is proposed. Three feature enhancement methods minimum entropy deconvolution， autoregressive linear prediction and wavelet de-noising are used to enhance the fault impact characteristics of gear vibration signals， and then the cepstrum is used to extract the fault features. The effectiveness of the proposed method is verified by experiment.