<p>Standard accounts of concepts in philosophy and in cognitive psychology hold that we learn a target concept by learning the concepts that constitute it. If this is the only model of learning we have, then only concepts that are complex or structured can be learned. I argue for an alternative to this framework by presenting a proposal for how we learn some atomic concepts. I show that we can learn a concept through descriptions without thinking that the concepts employed in the descriptions constitute the target concept. I develop this idea by using the notion of a mental file and by employing a general externalist explanatory strategy. In the view I develop, some concepts are akin to files in that the descriptions associated with a given concept are means of properly individuating the extension of the concept, and they are also means of connecting a concept to others, but they do not constitute the concept and can be revised or abandoned. I show how this discussion connects to issues in conceptual engineering.</p>