Abstract: In this investigation, 61 female and 44 male college student subjects participated in three experiments evaluating a central construct in neurolinguistic programming (NLP), that of the primary representational system. Authors Bandler and Grinder (1976) postulate that experience is encoded for storage and retrieval via hypothetical cognitive mechanisms called representational systems, and that individuals manifest a preference for one of three sensory-analogue systems (visual, auditory, or kinesthetic); this preferred system is the primary representational system. The first experiment dealt with the proposition that representational preferences can be reliably determined. Based on Bandler and Grinder's assertion that eye movements in particular directions, while subjects are generating responses to questions, reflect the use of particular systems, a structured interview was developed and used to assess the preference of subjects on two occasions (separated by a one-week interval, to permit assessment of temporal stability). Although several approaches to interpreting the data were employed to deal with areas of ambiguity in the construct, results consistently indicated confounding of observed preferences with method of measurement. In addition, very few subjects exhibited more than marginal preferences, no more than a modest degree of temporal stability was found, and discriminant validity was lacking. The second experiment evaluated the proposition that experience congruent with representational preference would have special salience or impact. Subjects evaluated audiotaped vignettes in which commonplace pleasant experiences were presented in visually- oriented, aurally-oriented, and kinesthetically-oriented versions. Over three stimulus experiences and five rating dimensions, no relationship was found between representational preferences and subjects' preferences among versions. The final experiment evaluated the hypothesis that clients' perceptions of therapists as trustworthy and effective are enhanced when therapist language matches clients' representational preferences, with respect to sensory referents. Subjects evaluated therapists heard in what were presented as excerpts from actual sessions, but were in fact staged interactions varying with respect to linguistic matching between therapist and client as well as matching between therapist language and subjects' representational preferences. In contrast to the predicted outcome, therapists who matched clients' language were evaluated as less trustworthy and effective, as were, independently, therapists whose language matched the primary representational system of the evaluating subject. The results illustrate the problems posed for psychotherapy when models proliferate in the absence of empirical evaluation.