This experiment used five index languages to form nine different index files to the same collection of documents. Each index term, in each language, was treated in two ways: as an independent English term, and as a coded entry in a faceted-type thesaurus which specified relationships between terms. The performance effect of the thesaurus was then tested as part of the question analysis procedure by enlarging the question terminology with and without use of the thesaurus. Six levels of search statements were thus formed for each question, which was then searched at all six levels over each of the nine index files. The results were analyzed to compare documents retrieved by the system with documents judged relevant by the users. The analysis indicates that: (1) enumeration of all relevant search concepts is tedious but necessary; (2) thesauri, as constructed today, do not seem to be powerful tools for term analysis; (3) even a very elaborate question analysis cannot avoid the problem of the inverse relationship between relevance and recall; and (4) information systems need to have designed into their operations elaborate procedures and a large capacity in terms of people, cost, time, etc., for the handling of questions. The author concludes that the whole subject of question handling in information retrieval needs additional and thorough investigation. (JW)