Presents a novel and robust method for leaf-position verification with a multileaf collimator (MLC). On the portal image associated with an MLC-generated treatment field, all true treatment-held-edge lines are either parallel or perpendicular to each other. This unique feature of an MLC treatment field has been fully exploited by the authors' method. Employing a Hough-type transformation as an edge-line-orientation detector and a chamfer-matching method, the authors can find the best matching parameters (including translation, rotation and scaling) adaptively between a prescribed MLC leaf configuration and the actual treatment-held edges generated by the MLC system. This works even if the portal image is partially corrupted by noise or covered by compact bony structures. Comparing these parameters with clinically accepted tolerances, the authors can make a "go-or-no-go" decision quickly.