Objective To determine the effects of a behavioural intervention for prevention of HIV and sexually transmitted diseases that identified, trained, and engaged leaders of Roma (Gypsy) men's social networks to counsel their own network members. Design A two arm randomised controlled trial. Setting A disadvantaged, impoverished Roma settlement in Bulgaria. Participants 286 Roma men from 52 social networks recruited in the community . Intervention At baseline, all participants were assessed for HIV risk behaviour, tested and treated for sexually transmitted diseases, counselled in risk reduction, and randomised to intervention or control groups. Network leaders learnt how to counsel their social network members on risk prevention. Networks were followed up three and 12 months after the intervention to determine evidence of risk reduction. Main outcome measure Occurrence of unprotected intercourse during the three months before each assessment. Results Reported prevalence of unprotected intercourse in the intervention group fell more than in control group (from 81% and 80%, respectively, at baseline to 65% and 75% at three months and 71% and 86% at 12 months). Changes were more pronounced among men with casual partners. Effects remained strong at long term follow-up, consistent with changes in risk reduction norms in the social network. Other measures of risk reduction corroborated the intervention's effects. Conclusions Endorsement and advice on HIV prevention from the leader of a social network produces well maintained change in the reported sexual practices in members of that network. This model has particular relevance for health interventions in populations such as Roma who may be distrustful of outsiders. Trial registration Clinical Trials NCT00310973.