This paper presents a robust, reliable iris location system for close-up, grey scale images of a single eye. The system is meant as a bootstrap or recovery module for automated iris tracking within medical applications. We model the iris contour with an active ellipse, sensitive to intensity gradients across its perimeter. In this way, we avoid modelling the noisy appearance of the iris (e.g. corneal reflections). The iris-sclera intensity transition is modelled at two spatial scales with Petrou-Kittler optimal ramp filters. The optimal ellipse is identified by a simulated annealing algorithm tuned to the problem characteristics. The system performed accurately and robustly with 327 real images against substantial occlusion levels and varying image quality, subject, eye shape and skin colour.