A randomised controlled trial studied the effect of an educational visit on benzodiazepine prescribing. An approximately representative sample of 286 general practitioners was allocated to an intervention or a control group. Rates of benzodiazepine prescriptions were derived from two comprehensive self-report surveys seven months apart. Two months after the first survey the intervention group received an educational visit and supporting material from a doctor or pharmacist, ostensibly unconnected with the surveys. The overall benzodiazepine prescribing rate fell by 23.7 per cent from the first to the second surveys, from 4.93 to 3.76 prescriptions per 100 encounters (P < 0.001). Anxiety and insomnia diagnosis rates also declined from 4.68 to 3.76 per 100 encounters (19.7 per cent). After adjusting for confounders, there was a differential downward trend in prescriptions per diagnosis of insomnia but not to a statistical level. The same was true of initial prescriptions per insomnia diagnosis. In a subsidiary analysis selecting only new insomnia diagnoses, the intervention had a strong effect in reducing initial prescriptions (odds ratio 0.18, 95 per cent confidence interval 0.04 to 0.73). No effect was seen on prescribing for anxiety diagnoses. Educational practice visiting for benzodiazepine prescribing in anxiety, as we conducted it, is not justified in an unselected population of general practitioners. Specific education on prescribing for insomnia is probably useful. Our interpretation of the reduction in benzodiazepine prescribing is that probably there was an effect from self-monitoring alone which overwhelmed a main-analysis intervention effect. Retrospective diagnosis may also have obscured a real intervention effect.