To help users find their way around the large number of titles available, public libraries organize the collection by placing titles on shelves by genre, by adding classification codes in their online catalogs, and by pictograms on book covers. However, users may have different perceptions of how titles are to be grouped in genres or categories. Public library collections will be more accessible when the way their collection is organized matches these users' perceptions. The authors show how user perceptions can be derived from library loan transaction data. To overcome the particular methodological issues in using such data, a method of ultrametric trees and latent class analysis is developed to determine segments of users and how they categorize the collection. Results show clear user categorizations that in particular respects differ from conventional public library categorizations. © 2005 Elsevier Inc. All rights reserved.