Twelve subjects from two job categories, sales engineers and programmer analysts, used an REL ENGLISH database to answer a set of questions. These questions were designed to require successively more complex interactions. The database contained Hewlett-Packard's Condensed Order Records, which were pertinent to the jobs of the sales engineers.All of the subjects were given a battery of cognitive tests measuring cognitive style and pattern extrapolation skills prior to using the database. They also received a brief training session on the structure of the database.Analysis of the subjects interactions with the REL ENGLISH database, particularly analysis of the errors made, showed: first, that cognitive style is significantly correlated with the number of questions successfully completed; second, that while sales engineers were able to access all levels of the hierarchy in the database, programmer analysts had significantly more difficulty accessing data from higher levels than they did with data from the same or lower levels than the standard, entry level; and third, that programmer analysts had less difficulty with the fixed-format, programming-language-like features of REL ENGLISH, while sales engineers has less difficulty with the free-format, English-like features of REL ENGLISH.These findings suggest that quasi-natural language database interfaces are appropriate for nonprogrammers who have a field-independent cognitive style and who already are domain experts in the area covered by the database.