Vignette and laboratory experiments suggest that negative reactions to people with mental illness are a direct consequence of their symptomatic behavior, but because of their poor external validity, these studies cannot tell us whether widespread negative public reactions to people with mental illness actually result from observation of symptomatic behavior. Focusing on perceived danger, we use a large national survey to test the "behavior hypothesis" in the general population. We reason that, if this hypothesis is correct, contact with people with mental illnesses should be associated with more perceived danger, and exposure to threat or harm should mediate this association. On the contrary, respondents with more personal and impersonal contact perceive people with mental illness to be less dangerous. Exposure to threat is more common among people with more contact, but this exposure explains very little of the variance in perceived danger. These findings do not support the conclusion that public fear of people with mental illness is due to the observation of violent behavior. (Contains 2 tables.)