Gaze tracking can be a powerful tool to control a computer or robotic limb. There are many gaze tracking technologies, but vision based gaze trackers, which work off of a standard web cam, have the potential to be quickly accessible. This report assumes that noisy estimates of the head position and gaze direction are known, and then presents a method to find a more stable estimate of the user's gaze on a table plane. The method works by combining estimates of each eye's gaze intersection with the viewing plane. The basic combination achieves fair precision but poor accuracy, so the report describes a simple way to improve accuracy while maintaining good aspects of the system. The system achieves an accuracy of 6.9 cm in the X direction and 5.7 cm in the Y direction for a gaze point experiment.