An on-line automatic diagnostic technology is presented in this paper, based on the simplified vibration model of early scratch defects of rolling element bearings and the analysis of vibration characteristics of early scratch defects. According to characteristics of periodic high frequency natural vibration excited by early scratch defects of the bearing, high frequency resonance technique (HFRT) is adopted for on-line diagnosing in the paper. An algorithm on peak detection of the vibrational energy is used to catch the main frequency of high frequency resonance automatically, denoising technology to improve the signal-to-noise ratio, and automatic identification technology to evaluate defect site of the tested bearing. An on-line automatic diagnostic system for early scratch defects of rolling element bearings, developed based on the above principles, can effectively identify early scratches on outer race, inner race and ball.