OBJECTIVE: Changing behaviour is an important method for preventing HIV infection. We examined why a community randomized trial of a behavioural intervention found no significant effect of this on HIV incidence in rural Uganda. DESIGN: An individual-level analysis of a community randomized trial. METHODS: All sexually active, initially HIV-seronegative individuals with data on sexual behaviour were included (1558 men and 1836 women). Uptake of the intervention was measured using self-reported attendance at meetings, videos, dramas, and interactions with community educators in the past year. Sexual behaviour was assessed using self-reported condom use and the number of sexual partners in the past year. RESULTS: Overall, 81% of individuals in the intervention communities and 9% in the comparison communities reported attending at least one of the intervention activities in the past year. Attendance was lower in women, in those aged 55 years or older, and in the widowed. There was a lower HIV incidence in those who reported attending at least one intervention activity compared with those who attended none, and in women this effect was statistically significant (in women, adjusted rate ratio 0.41, 95% CI 0.19-0.89, P = 0.024; in men, adjusted rate ratio 0.66, 95% CI 0.25-1.79, P = 0.42). Reported behaviour change did not differ markedly between those who did and did not report attending any intervention activities. CONCLUSION: Although the intervention had no significant benefit in the communities as a whole, it resulted in a reduced risk of HIV acquisition in women who attended it. The methodological implications for future trials are discussed.