This article reviews two aspects of human learning: (1) people draw inferences that appear to rely on hierarchical conceptual representations; (2) some categories are much easier to learn than others given the same number of exemplars, and some categories remain difficult despite extensive training. Both of these results are difficult to reconcile with a learning and categorization system that operates only on specific exemplars. More generally, the article argues that specifying the empirical phenomena that a radical exemplar does not predict would aid in clarifying the radical exemplar proposal. [For Ben Ambridge's "Against Stored Abstractions: A Radical Exemplar Model of Language Acquisition," see EJ1269951.]