In the actual production line, bearings groove curvature radius of the same type is detected online according to different size deviation. To solve this problem, an online detection system of bearings groove curvature radius based on machine vision was designed to replace manual detection. The method combining CCD camera with image processing technology of MATLAB was used to make noncontact measurement of the parts and to preprocess the collected parts images. By comparing various edge detection algorithms, Canny algorithm was adopted to obtain the edge of pixel precision for the bearing. The circular Hough transform was used to detect the circle features with arcs, and the circle center coordinates and radius values of this arc were calculated. The experimental results show that the measurement accuracy of the system can reach 0.5 μm and the standard deviation of the measurement is less than 2.5 μm, it meets the requirements of industrial testing.