In an experimental controlled trial, prescribing records were obtained from the Medicaid Management Information Systems (MMIS) of four states for all physicians participating in the Medicaid programs of those states. Three categories of drugs that are commonly misprescribed were identified, and moderate to high prescribers of these drugs were identified from the MMIS data set. These physicians were then randomly divided into three groups. One group received no intervention, the second group received an innovative series of print materials urging appropriate drug use, and a third group received the print materials and was visited by consultant pharmacists to discuss the drugs in question. Our experience suggests that use of Medicaid prescribing data can be an efficient and accurate way of conducting large-scale surveillance of misprescribing, and of targeting interventions that can improve such suboptimal drug utilization. Use of the same data set in a follow-up period can monitor the effectiveness of each mode of intervention and will measure the degree of behavior change for each physician.