Participants first became familiar with an image showing moderate symptoms of the skin cancer melanoma. In a generalization test, they indicated whether images showing more and less pronounced symptoms were "like the original." Some groups (cancer context) were told that the images depicted melanoma and that the disease is deadly unless detected early. Control groups were not told what the images depicted. For control groups, generalization gradients were fairly typical of what is normally reported in the generalization literature, but for cancer context groups, gradients were shifted such that highly symptomatic moles were identified as "like the original" more than normal and subtly symptomatic ones were endorsed less than normal. These results may have implications for melanoma education efforts and, more generally, illustrate the possible importance of studying interactions between verbal behavior and primary stimulus control.