In this paper we present the results of two experiments designed to understand how physics students' learning of the concept of refraction is influenced by the cognitive phenomenon of "specificity." In both experiments participants learned why light bends as it travels from one optical medium to another with an analogy made to a car driving from paved road into mud and vice versa. They then learned how to qualitatively draw the direction of refracted light rays with an example of a glass prism. One group learned with a rectangular prism example while a second group learned with a triangular prism example. In a transfer test, the participants revealed how, even when they seemed able to implement the refraction concept, their responses were biased by the example they had seen. Participants frequently violated the refraction principle they had just learned (reversing the bend direction) in order to make sure their response matched the surface features of their learning example. This tended to happen when their test question looked superficially similar to their learning example. We discuss the implications of these results for physics instruction.