This paper describes an attempt to facilitate students' learning of scientific concepts by using detectors that take as input physical information and output an instantiation of the concept. The principle hypothesis was that students would have a better understanding of the concept of image if they were taught to use a simplified, runnable model of their own visual detectors, than if they were taught using a more traditional approach. To test this hypothesis, two almost identical versions of a curriculum in geometrical optics were created, one version used a mechanistic, interpretive eye model, while the other treated the eye as a passive receiver of light. Results indicate that students who were taught a runnable model of visual perception exhibited a better understanding of the difficult relationship between an observer and a virtual image, were better able to identify the location of an image in a range of real-world optical situations, and were less likely to think of it as located on the surface of a mirror or lens. It was concluded that significant gains are made by teaching students a transparent-detector model of image formation rather than a more traditional geometrical definition of image formation. (JRH)