In option searches, a user seeks to locate an ideal option (e.g. a flight, restaurant, book, etc.) from a set of n such options. The aim of this paper is to provide a solid mathematical basis for optimizing presentation length in such searches. The paper develops an information theoretic model that takes into account the user’s ability to discern among options as well as their a priori preference. The developed model makes definite predictions about what clusterings of a user query are more or less informative based on measures of information gain. Users are offered descriptions of such clusters as the basis for subsequent refinement steps in a drill-down dialogue to locate the best option. We have implemented an initial system that performs reasonably well on moderately large data sets and gives intuitively appealing results. The system is in the process of being integrated into a natural language interface system for end-user evaluation.