A peer-led HIV prevention initiative, based on a diffusion of innovation model, was developed for gay men attending gyms in central London. Peer educators were recruited from people who used the gym regularly according to standard selection criteria. After initial training, peer educators agreed to talk to gay men at their gym about HIV prevention, focusing on sexual risk and steroid injecting behaviour. Outcome evaluation revealed that the peer education programme had no significant impact on the risk behaviours of gay men using the gyms. Process evaluation, based on interviews with peer educators, the health promotion team and gym managers threw light on this finding. While it was feasible to set up a peer education programme among gay men in central London gyms, attrition was an important factor. Only one in five potential peer educators initially identified remained with the project throughout, thus limiting the potential for diffusion. Those who did work as peer educators reported barriers to communication within the gyms further limiting the extent to which diffusion occurred. In fact, it appears that the critical mass required for diffusion was never established. This could explain why the intervention had no significant impact on gay men's risk behaviours. A person-time analysis demonstrated that the peer education programme required a substantial input from the health promotion team, equivalent to one team member devoting 2.5 days a week to recruit, train and support peer educators over 18 months. Peer education should not, therefore, be viewed as a low-cost approach to prevention. Many of the insights gained through this process evaluation can inform others planning peer education programmes in other settings.