This article describes 5 experiments investigating the role of prior knowledge in incidental category learning. Experiments 1 to 3 showed that prior knowledge improved learning only if the categories in a given set were related to contrasting themes; there was no consistent knowledge effect if the categories were related to the same theme. Experiments 4 and 5 showed that diagnostic verbal labels facilitated the learning of non-thematic categories but provided no additional benefit when the categories were already related to contrasting themes. In terms of the category invention framework proposed by Clapper (2007), these results imply that prior knowledge provides an effective cue for discovering separate categories, as well as helping people segregate the features of those different categories in memory and so improving their recall. The relevance of these results to other types of category learning tasks is discussed.