Research on category-based induction has shown that typical exemplars promote stronger inferences about categories than atypical exemplars. This work has been largely confined to categories that are typified by the central tendency of category members. Does the typicality effect apply to categories for which the ideal category member is considered most typical? In experiments with natural and artificial categories, we obtained ratings of typicality and inductive strength for ideal and average exemplars. Inductive strength was greatest for the average exemplars, regardless of whether the average or the ideal was rated more typical. These results suggest that people perform category-based induction by a process akin to statistical inference from one of multiple simultaneously-held category representations, rather than comparison to a single canonical representation.