When people search the medical literature, they often are overwhelmed by the large number of documents retrieved. Many systems try to solve this problem by helping the user formulate a more specific search strategy. However, when users do not have a more specific question, they need tools to help them explore and understand the results, rather than to eliminate a portion of those results. This paper describes an approach that addresses this need by automatically grouping the results of a broad search into meaningful categories based on the user's query. This approach combines the main benefit of clustering techniques with the main benefit of classification techniques by taking advantage of the domain knowledge present in the UMLS. I present a preliminary evaluation that demonstrates that a categorization produced by this approach corresponds reasonably well to a physician's categorization.