A method based on the morphological gradient(MG)filter was proposed for feature extraction of gear fault signal.A comparison was made bteween the conventional envelope demodulation and the morphological close method,and the latter method was applied to the simulated signals and the measured gear fault signals interfered by low-frequency harmonic signals and random noise.The results demonstrate that the MG filter can effectively suppress the noise and clearly highlight the impulse feature of the original fault signals and be entirely free from interferences of the low frequency harmonic.Moreover,the computation cost of the MG filter is much less.So,the MG edmodulation filter provides an efficient approach to feature extraction for gear fault diagnoses.