In view of the flaws of modeling difficulty and sample training difficulty found in most current mechanical fault diagnosis technologies, a mechanical fault diagnosis technology, which is based on continuous wavelet transform time-frequency analysis and maximum inter-class threshold image segmentation, has thus been proposed. By adopting a time-frequency analysis technology based on continuous wavelet transform, the measured time-domain vibration signal is transformed to the frequency-domain space, and then represented in the form of vibration thermal diagram by using the visualization method. The area calculation of each frequency component distribution can be realized by preprocessing the regional time-frequency features after an image enhancement, followed by an analysis of the vibration signal in the way of characteristic area ratio. The experimental results show that the proposed method is characterized with a fast processing speed, an improved process transparency, and a good adaptability.