As automation of office functions becomes more widespread and longstanding, large on-line databases of documents will commonly be accessed by personnel unschooled in database logic. It is necessary, therefore, to design query languages for document retrieval that are intuitive, simple, and fast. It is likely that effective query languages will be those that systematically provide the user with powerful memory cues, with highly interactive structures guiding the formation of queries, and with a very simple and learnable syntax. No one query language will suit every potential user; a means of optimizing the design of query languages, therefore, is to fit each language to the requirements of a specific population of users. At Exxon Office Systems we have, in the past year, designed two query languages, one aimed at occasional users, the other at regular users. These languages are attempts to provide convenient document management facilities of the kind alluded to above. A study was implemented to test the learnability and usability of these languages, using flat file search as a baseline. Sessions involved retrieval by subjects of documents from large (500 member) artificially generated databases of documents; subjects were run over multiple sessions. The raw data consisted of full keystroke capture of each session. By replacing the database mid-experiment, effects of acquisition of the query languages were separated from effects of memorization of the database. Learnability and usability were evaluated by statistical analysis of learning curves, command usage, response times, and individual differences. Both query methods appeared far preferable to flat file search, especially as users gained facility with them. One method was more easy to use but less powerful–and appeared to be appropriate for casual users; the other method was slightly harder to use, but provides the regular user with a more flexible and powerful tool for document management. Another outcome of the study was a preliminary model of the document management process–a model whose parameters depend on features of the query language, the database, and the population of users. By observing how features of a query language affect the parameters of the model it should be possible to refine the language so as to improve its effectiveness for a given population of users.