When a gear has a local fault, the vibration signals of gear may contain amplitude and phase modulation. Their spectrums contain meshing frequencies, harmonics, and coupling frequencies generated by modulation. According to the characteristics of vibration signals from the gears with faults, a feature extraction method based on the envelope and time-frequency image of S transformation was proposed. Firstly, vibration signals from gears with differ worn degrees were collected by fault simulation in an experimental setup. Then, envelopes were obtained by Hilbert transform from vibration signals and envelopes of time-frequency contour maps were achieved by S transformation. Finally the features were extracted by calculating statistic parameters based on the grey-scale matrixes (GLCMs) of the maps. The result shows that the proposed method can effectively extract gear fault feature.