The purpose of the research project was to compare three types of interfaces to online information retrieval systems (command-driven, menu-driven, and associative) in a fair manner avoiding favoring one over others. Since match, search and retrieval processes affect response time we wrote a single back end for those functions. To reduce the number of variables we prepared a single database and a standard set of 24 queries. The interfaces were tested by human subjects to determine which type of interface is most acceptable to novice users. The database used was a subset of MEDLINE file of the NLM. We created a minidatabase of references, with abstracts, dealing with topics (such as anorexia nervosa, tennis elbow and Health Maintenance Organizations, HMOs) that are familiar to the general public. The base included 34,000 references, an amount that would fit on a single disc on our VAX11/780. File building included generating indexes, term frequency lists and a stop-word list of prepositions, conjunctions, etc, and very high frequency words that were query words and synonyms (to avoid degrading response time).