Fault feature is always hidden by strong noise background in gear fault signal. Based on morphological wavelet de-noising, a novel method was proposed to extract gear fault feature. Morphological wavelet de-noising has a good performance in extracting morphological feature in signal. Firstly, the signal was decomposed by morphological wavelet. Secondly, detail coefficient in each level was processed using soft threshold de-noising. Finally, fault feature was extracted by reconstructing original signal. Simulation and experiment results showed that this method is effective in gear feature extraction. Morphological wavelet de-noising algorithm includes addition, subtraction, maximum and minimum operations, and does not involve multiplication and division. It is suitable for on-line monitor and gear fault diagnosis.