Use of active vision to automatically track a mandrel used for shear spinning is presented. The silhouette of the mandrel is imaged by a camera mounted on the tool magazine. The machine axes were actively controlled using an external computer with visual feedback. The visual feedback signal generates sub-pixel accuracy. The presented system eliminates parallax error and spherical aberration by actively positioning the camera axis to pass through the mandrel profile. Although visual feedback is used, the mandrel profile is recorded directly using the machine coordinates thereby avoiding the image plane to real world transformation. The recorded positions were further processed using a segment extraction algorithm. The actual mandrel data, the raw tracking data as well as the segment-extracted data were used to spin three components. Results obtained using 3-D coordinate measuring machine are presented to prove the accuracy of the tool path generation system.