A microcomputer system with a desktop scanner is employed for locating the landmarks of an X-ray cephalogram automatically. The digitized image of a cephalogram is processed to define a reference point for analysis and divided into several rectangular subimages (such as frontal, orbit, maxilla, mandible, etc.). These regions containing all the useful landmarks are then enhanced in a preprocessing stage. The pyramid method is used to reduce the resolution of image, and the edges are detected by appropriate edge detectors or the best-orientation edge detector. The curve of each edge is corrected elastically using prestored models. Finally, positions of landmarks are located immediately and the associated parameters are computed for diagnosis.