A distinction is made between two methods for representing concepts. Negatively defined concepts are influenced by other concepts at the same level of abstraction. Positively defined concepts do not depend on other concepts at the same level of abstraction for their characterization. Methods for manipulating a concept's representations and for identifying positively and negatively defined concepts are introduced. Positively defined concepts are empirically identified by a relatively large use of nondiagnostic features, and by better categorization performance for a concept's prototype than for a "caricature" of the concept. Negatively defined concepts are identified by minimal use of nondiagnostic features, and by better categorization performance for a caricature than a prototype. A concept is likely to be more positively than negative defined when: subjects are instnucted to create images for their concepts rather than find discriminating features, concepts are given unrelated labels, and when the category that is displayed alternates frequently between trials. The entire set of manipulations and measurements supports a graded distinction between positively and negatively defined concepts. The distinction is applied to current models of category learning, and a connectionist framework for interpreting the empirical results is presented.