To assess whether the spread of infection with HIV can be reduced by changes in behaviour among groups most at risk because of their sexual practices sexual behaviour was monitored among 1050 homosexual men tested for HIV infection at a genitourinary medicine clinic in west London from November 1984 to September 1987. Four cohorts, defined by date of presentation, were studied by questionnaire at their presentation, and blood samples were analysed. Between the first and last cohorts there was a considerable fall in the proportion reporting casual relationships (291/329 (88%) v 107/213 (50%] and high risk activities, such as anoreceptive intercourse with casual partners (262/291 (90%) v 74/106 (70%], with the greatest changes occurring before the government information campaign began in 1986. Nevertheless, half of the men in the last cohort studied reported having casual partners. Multiple logistic regression showed that behavioural risk factors for HIV infection most closely resembled those for hepatitis B and that previous sexually transmitted diseases (syphilis, hepatitis B, and anogenital herpes) were themselves independent risk factors. A history of syphilis ranked above anoreceptive intercourse as the strongest predictor of HIV infection. Actively bisexual men showed a much lower prevalence of HIV infection (3/57, 5%) than exclusively homosexual men (113/375, 30%). Sexual behaviour among homosexual men changed during the period studied, and the incidence of HIV infection fell, although more education programmes directed at homosexual men are needed to re-emphasise the dangers of infection.