This chapter reviews that research on human concepts can be divided into two general categories. One class includes formal models of categorization, in which the goal is to discover the learning algorithm and representational format for concepts. This work addresses the empirical aspects of learning concepts, in which a learner is exposed to a certain number of examples and must induce concept representations for the examples categories. The other class is the “knowledge-based” approach, in which the influence of general knowledge structures on concepts is investigated. It discusses that the issue is how one's “theories” about a domain or one's expectations influence the information learned about concepts and how that information is organized in memory. There is no well-developed mathematical analysis of knowledge structures and their effects, therefore, this work typically does not result in formal models. The chapter also describes Anderson's dismissal of theory-based approaches, linguistic categories, and ad-hoc categorization is an excellent example of the chasm between the two approaches to concepts. No matter how convincing an arguments might be for dismissing one of these approaches, neither of them by itself is likely to explain concept formation in full. However, saying that the two approaches are both necessary is not the same as integrating the two, and this is where much of the future work in this field needs to be focused.