In this paper, a vision-based inspection system that measures the dimensions of a ball stud is designed and implemented. The system acquires silhouetted images by backlighting and extracts the outlines of the nearly dichotomized images in subpixel accuracy. The sets of boundary data are modeled with reasonable geometric primitives and the parameters of the models are estimated in a manner that minimizes error. Jig-fixtures and servo systems for the inspection are also contrived. The system rotates an inspected object to recognize the objects in space not on a plane. The system moves the object vertically so that it may take several pictures of different parts of the object, resulting in improvement of measuring resolution. The performance of the system is evaluated by measurement of the dimensions of a standard ball, a standard cylinder, and a ball stud.