This thesis describes an investigation into the experimental accuracy of a close-up stereo vision system. Surfaces are measured from any orientation through the synergy of a Coordinate Measurement Machine (CMM) and a custom made stereo vision head. Calibration of the stereo cameras is achieved using bundle adjustment non-linear optimization, commonly used in photogrammetry. Point set registration techniques are used to calibrate the stereo head with respect to the CMM coordinate system. The fully calibrated stereo vision system is used to provide experimental accuracy results under ideal imagining conditions. The system is then applied to the practical problem of sheet metal strain analysis. This problem involves many non-ideal imaging components that affect system accuracy. Experimental results for this non-deal situation are provided in the form of three-dimensional strain distribution plots.