In information facilities such as libraries, finding documents that are relevant to a user query is difficult because of the indeterminism involved in the process by which documents are indexed, and the latitude users have in choosing terms to express a query on a particular topic. Reference librarians play an important support role in coping with this indeterminism, focusing user queries through an interactive dialog. Based on thirty detailed observations of user/librarian interactions obtained through a field experiment, we have developed a computational model designed to simulate the reference librarian. The consultation includes two phases. The first is handle search, where the user’s rough problem statement and a user stereotyping imposed by the librarian are used in determining the appropriate tools (handles). The second phase is document search, involving the search for documents within a chosen handle. We are collaborating with the university library for putting our model to use as an intelligent assistant for an online retrieval system.