The indexing and online searching of data bases have both conflicting and complementary structures, determined in part by the choice of indexing language. In the planning of a machine readable system, the Foundation Center, administrator of two data bases, initially chose a natural language for the lower cost, more rapid implementation, and greater efficiency than that of a controlled vocabulary. However, the input cost minimization afforded by the natural language alternative may result in cost shifting rather than reduction by increasing the cost of searching. Searchers also found the use of natural language undesirable due to the imprecision of the language, existence of many synonymous terms, and the difficulty of transfer of search strategies across systems. The decision was made to construct and initiate a controlled vocabulary index. The process of descriptor compilation and validation included rank ordering of present free-text descriptors' frequency, identification of synonyms, identification of relationships (e.g., broader term, narrower term), expert validation, classification of related trees, and construction of the thesaurus review process. (JVP)