An unmatched retrospective case control study was conducted to test the feasibility of investigating social and behavioural factors which may have contributed to recent HIV seroconversion in a group of homosexual men. Participants, recruited from a London sexually transmitted disease (STD) clinic, were sexually active and had had a negative HIV test with a subsequent test (positive (cases) or negative (controls)) within three to 15 months. Twenty cases and 22 controls were recruited between February and October 1995. There was no difference between cases and controls in: the number of regular or casual sexual partners, the proportion who were unaware of their regular partners' serostatus (cases 60%, controls 59%), or the proportion who had known HIV-positive regular partners (cases 20%, controls 23%). A significant difference in sexual behaviour was found only when the HIV status of partners, if known, was taken into account: cases were more likely than controls to have had unprotected receptive anal intercourse with a partner not known to be HIV-negative (OR = 5.5, CI = 1.15-29.50). Fifty per cent of the cases and 27% of the controls acquired acute STDs between the two HIV tests. All participants achieved high self-efficacy scores, but the controls believed their peers placed a greater value on safer sex. Cases cited emotional issues and the use of drugs and alcohol as contributing to their seroconversion, whereas controls cited a commitment to safer sex and the avoidance of high-risk situations as contributing to their remaining HIV-negative. The results illustrate the importance of acknowledging the concept of 'negotiated safety' in studies of sexual behaviour; seroconversion was only associated with unprotected sex with a partner not known to be HIV-negative. Despite high self-efficacy scores, indicating the skills to negotiate safer sex, high levels of unsafe anal intercourse were reported. Differences between cases and controls included the importance of safer sex, periods of emotional vulnerability, influence of peers and the appropriate use of condoms. There is a need for these results to be confirmed in a larger and more powerful study.