Abstract: A method was proposed for fault feature extraction based on the gray-scale hit-or-miss transform. Several target waveforms are chosen from the analyzed signal, and then hit-or-miss structure element pairs could be thereby calculated. Pattern matching is carried out at every sample point, and higher output is obtained when fault features occur. As the method was used to analyze the rubimpact signal of a driving motor and the impulse signal of a rotator test-bed in an oxygen manufactory, fault features were successfully extracted. The method is close to artificial thinking and easy to understand, so it is an effective method to extract fault features from a time domain waveform straightly.