We perform a linguistic analysis of documents during indexing for information retrieval. By eliminating index terms that occur only in subordinate clauses, index size is reduced by approximately 30% without adversely affecting precision or recall. These results hold for two corpora: a sample of the world wide web and an electronic encyclopedia.