3D graphic scenes are only correctly rendered for one viewpoint. Without laborious calibration, however, observers seldom view the monitor from this viewpoint. Even in visual experiments using headrests, inter-subject variability in head-size and eye position result in many subjects viewing the display "off-axis", producing well known distortions in perceptual judgments. The goal is to correctly render graphic displays for the application/experiment based on a simple set of perceptual judgments made by the user. We have two approaches. Our first approach uses point matches between points on a transparency and 3D haptic points that the user makes with the Phantom device. We use well-known calibration techniques from Computer Vision to estimate the transformation matrix between the mirror and the monitor and also the position of the eye. This method requires the presence of a 3D calibrated object (Phantom, in our case). Our second method uses the same transparency and user-adjustable points on the monitor to derive a transformation matrix between mirror and monitor, as well as the position of the eye. This method does not require the presence of a calibrated object and hence is more generally applicable.