Surface quality of bearing ball is one of important factors to improve service performance. It is difficult to detect bearing ball surface defects with high precision and low cost by conventional methods. In this paper one new classification method based on optical fiber sensing technology was put forward from the longitudinal section angle. Inspection mechanism of Y-type and coaxial fiber sensor were analyzed, the mathematical models of the two fiber sensors were established and numerical simulations were realized by Matlab software. Influence of optical fiber parameters on light intensity modulation characteristic regulation was studied. Light intensity modulation characteristic experiments were done with specimens of different reflectivity and processing methods. The results show optical fiber sensing technique is an effective method to separate different surface of steel ball with extracting of two parameters displacement and surface reflectivity.