An automatic partial-pose measurement system for robot calibration, called the vision-based automatic theodolite (VBAT), is described. It uses low-resolution rotation stages and resolution enhancement from a vision system to determine the line of sight to a spherical illuminated target. Automatic tracking, focusing, and centering provide the calibration system with speed, reliability, and repeatability. It is proposed that such a measurement prototype may be constructed without expensive precision machining, provided it is calibrated before use. Simulations demonstrate that such a system may be identified by precalibration of the vision parameters, followed by calibration of the rotation stage parameters. All parameters are determined from measurements using a coordinate measuring machine. The parameter identification is formulated as a nonlinear regression problem. Verification of the VBAT calibration experiment shows that the identified model can predict the target position to a root mean square miss-distance of 0.2 mm.>