In analyzing a university-based program to educate physicians about proper medication use, we sought to measure whether physician background characteristics and the quality or number of educational exposures influenced the rate of relinquishment of inappropriate prescribing. A sample of 435 doctors was randomized to control and experimental groups; interventions consisted of printed educational materials and face-to-face visits by clinical pharmacists. The program sought to reduce inappropriate use of three drug categories: propoxyphene, peripheral/cerebral vasodilators, and cephalexin. Outcome data included the total volume (tablets/capsules) of these drugs prescribed through Medicaid by each study physician 9 months before and after the program. We estimated average changes in prescribing levels by experimental and control physicians within each physician subgroup (e.g., board-certified versus uncertified), adjusting for prescribing level in the same 9 months of the previous year. The results indicated that the rate of prescribing change was independent of most physician background characteristics studied, including age, board certification, specialty, rural versus urban practice, intensity of previous target drug use, and size of Medicaid practice. Experimental effects were highly significant (-9% to -20%, P less than 0.025) in 11 of 14 physician subgroups. The presence of a follow-up reinforcement visit was a strong independent predictor of prescribing change (P less than 0.05). An increase from one visit to two visits was associated with an approximate doubling of the size of the program effect. However, total exposure time was not related to changes in prescribing behavior. These findings document that face to face education can be effective in improving the prescribing practices of a wide variety of physicians, and that brevity, repetition, and reinforcement of recommended practices are important components in the design of such programs.