The meshing impact caused by backlash has great effect on the stability and reliability of gear transmission.To acquire the impact component from the fault signals of gear system,a feature extraction method based on Gammatone filters is proposed.By using signal processing function of human basilar membrane in the complex speech recognition process for reference,the Gammatone filters is used for time-frequency analysis of fault signals according to their similarity in impulse response and amplitude-frequency,and further to extract the impact feature.Research results show the time-frequency analysis ability of Gammatone filters to the non-stationary transient vibration signals of the gear system and the effective feature extraction ability of the proposed method.