The debate on the nature of concepts has been very lively in Analytic philosophy for the last two decades. Concepts inherit some traditional issues from the philosophy of language, such as the nature of reference and the compositionality requirement, but also lie at the intersection between philosophy and cognitive psychology. In his latest book Hume Variations, Jerry Fodor frames the contemporary debate in terms of an opposition between Pragmatist theories, which describe concepts as categorization capacities, and Cartesian theories, according to which concepts are representations of objects and properties. According to Fodor's definition, the category of conceptual pragmatists is both large and variegated, including conceptual role semantics, but also prototype theories, exemplar theories, theory-theories and, in general, any psychological model that has been discussed so far in cognitive psychology. Amomg Atomists, Fodor enlists William of Ockham, David Hume and himself. According to Pragmatism, concepts are structured entities, carrying the information needed for categorization and inferential tasks. Atomism is the view that the components of thoughts are simple, unstructured symbols, and their unique function is to stand for some (generally) extra-mental entity. Thus, the opposition mirrors the long-standing debate between Fregean theories and direct-reference theories in the philosophy of language. I argue that the opposition is spurious when applied to theories of concepts - though it was genuine for theories of meaning. Concepts can be, and should be, both representations and capacities, once the notion of capacity is spelled out without ambiguities. Capacities need not be reduced to their behavioural manifestations. In fact, psychologists today tend to assume that capacities are represented in the mind-brain, they take the form of encoded knowledge - take, for example, Chomsky's account of the faculty of language. In this sense, there is no genuine theoretical choice to make between capacities and representations. Rather, the genuine issue about concepts is what kind of representation they are, and whether we should go on to assume that a single unified model will be able to account for the variety of concepts.