Many existing information access systems deliver ranked lists of documents in response to users' queries. Some systems also endeavor to represent some of the various types of relationships that can exist between documents. However, few systems provide effective mechanisms to help users discover useful information within the set of retrieved documents. In this paper, we present a question-driven approach to delivering retrieved documents in an attempt to organize them in a way closer to the user's mental representation of the expected answer. In our purposed approach, retrieved documents are dynamically classified into categories; an appropriate classification scheme is selected by a user on the basis of their own understanding of the information need. Experimental results show that users are more satisfied with such a directed categorization than with a list of retrieved documents.