The complexity of corneal refraction modeling complicates geometrical analysis of gaze tracking systems. The task can be accomplished in different ways by means of different approximations. Depending on the assumptions made, errors can arise and the accuracy of trackers can end up determined to an excessive degree by the hardware setup and subject parameters. In this study, keeping the hardware and eye model the same, we compare different methods for evaluation of corneal refraction. Our results show that the approximated methods are sensitive to the calibration of the system and less robust to noise. The nonapproximated method removes any dependency on setup and allows clear differentiation between hardware and subject's calibration.