To determine the effect of two levels of educational intervention on benzodiazepine-prescribing behavior in an elderly population in a controlled prepaid group practice (PPGP) setting, we designed a prospective controlled trial, with six-month follow-up. Our setting was a 270,000 member group-model PPGP in Colorado, from 1990 to 1991. Participants included 91 physicians, 62 men and 29 women; median age was 38.7 years. Group 1 received a one-on-one educational presentation by a clinical pharmacist, written educational materials, a brief follow-up visit, and feedback with recommendations. Group 2 received only a face-to-face presentation, given to departmental groups, as well as the same written educational materials used in group 1. Controls received no intervention. Our primary outcome measure was the benzodiazepine "on/off" status of the elderly PPGP members. The secondary outcome measure was the median change (preintervention minus postintervention) in a standardized amount of benzodiazepines prescribed per physician. Logistic regression analysis failed to show a significant effect on postintervention benzodiazepine on/off status between study groups, when controlling for preintervention on/off status, PPGP-member age, PPGP-member gender, and all possible interactions. Analysis of variance failed to demonstrate an effect of either intervention on the median change in standardized amount of benzodiazepines prescribed per physician, with groups 1, 2, and controls yielding values of -278 (range: -4,137, 2,844), -330 (-1,531, 1,358), and -541 (range: -3,716, 2,185), respectively. We conclude that strategies effective in changing physician prescribing behavior in other settings may not be effective in a PPGP setting with benzodiazepines in the elderly as the target for change.