The work aims to propose a non-contact measurement method based on machine vision, in order to improve the testing efficiency with respect to the situation that the traditional testing method cannot meet the testing requirements as the function features of complex surface part are diverse and the structure size takes on spatial distribution. The image data was collected with CCD camera to analyze and process the images, so as to obtain the sub pixel edge contour of a circle. The parameter equation of circle was obtained based on circle fitting by using the least square method. Lastly, the pixel distance could be calculated with the geometric distance formula. Pixel equivalent was got by system calibration and the actual distance between circle and circle could be obtained based on pixel equivalent. The testing algorithm of sub pixel edge of circle by the least square fitting was stable with good anti-noise performance and 0.001 pixel resolution. The proposed method is correct, convenient and effective when it is used to measure the part size.