This paper describes how computer vision is used to sense the pose (position and orientation) relationship between a robot tool and workpiece. This method is more direct than conventional techniques based on kinematic models and joint feedback. The approach is implemented by processing workpiece features and a geometric target mounted near the tool-tip. Precisely mounted cameras and a priori knowledge of the camera pose are not required. The paper emphasizes system calibration, single-view monocular vision, and control strategies. Finally, the method is implemented on a high performance industrial robot which illustrates how the powerful sensing capabilities of computer vision can improve the adaptability, flexibility, and cost of many robot systems. Once the proposed methodology is fully implemented and matured, the number of potential applications is enormous.