Schizophrenia is a chronic brain disorder that affects about 1.1% of the adult US population annually. Hallucinations, delusions, and impaired reality testing are prominent symptoms of the disorder. Modeling these symptoms is difficult because it is unclear how to assess impaired reality testing in animals. Animals cannot discuss their beliefs; however, a century of learning experiments has shown us that they, like us, construct complex internal representations of their world. Presumably, these representations can become confused with reality for animals in much the same way that they do for schizophrenic patients. Indeed, there is evidence from studies of Pavlovian conditioning that this happens even in normal animals. For example, early in training a cue that has been paired with reward elicits a highly realistic, sensory representation of that reward, which is to some extent indistinguishable from reality. With further training, this sensory hallucination of reward is replaced by a more abstract representation, termed a reward expectancy. Reward expectancies reflect the sensory and other qualities of the impending reward but are distinguishable from the actual reward. Notably, the hallucinatory representations depend on subcortical regions, such as amygdala, whereas reward expectancies require the progressive involvement of prefrontal areas, such as orbitofrontal cortex. Abnormal prefrontal function is associated with schizophrenia; impaired reality testing may result from a failure of the normal shift from highly realistic, sensory representations to more abstract, prefrontal expectancies. The Pavlovian procedures discussed here could be applied to animal models and schizophrenic patients to test this hypothesis.