Taking the measured gearbox vibration signals as the analysis object,the fault feature of gear pitting is extracted. The Akaike Information Ceriterion（ AIC） is applied to evaluate the optimum Auto- regressive（ AR） model order,and then this AR model is used to pre- process the vibration signals to eliminate the linearly predictable stationary part. The Maximum Correlated Kurtosis Deconvolution（ MCKD） is applied to enhance the impact component. The fault characteristic could be identified by analyzing the envelope spectrum.And the AR- MCKD method is applied to analyze the changing trend of the vibration signals. The results proved the efficiency of the AR- MCKD method in the gear fault feature extraction and the changing procedure of the envelope spectrum of gear pitting fault.