The automated or semiautomated control of microrobots for targeted drug delivery, retinal surgeries, and other ophthalmic procedures has the potential to significantly augment the capabilities of human surgeons. For accurate magnetic control, the position of the microrobots is required. In this paper, we extract the intraocular projection mapping and use it in the established CAD-model-based pose-estimation framework. Our algorithm requires no focus information. This is the first work that treats a complicated refractive imaging system, such as the eye, in a model-based localization framework. We design a model-eye chamber with human-like optical elements in which we control a microrobot and estimate its position using monocular vision.