OBJECTIVES: To develop methods to maximize the accuracy of reporting HIV risk behaviours in a general population survey. We assessed the feasibility of using a computer-assisted self-completion interview (CASI) in comparison with pen-and-paper self-completion interview (PAPI). DESIGN: A probability sample survey of residents aged 16-44 years in Britain, with alternate assignment of addresses to interview by CASI (462) or PAPI (439). METHODS: Personal interviews exploring demographic and sexual behaviour variables. Principal outcome measures were the impact of CASI in relation to PAPI on data quality and rates of reporting a range of behaviours. RESULTS: A total of 901 interviews were completed; 829 individuals were eligible for and accepted the self-completion module. Internal consistency of data items was greater with CASI than PAPI and item non-response was lower. Overall, there was no significant difference in rates of reporting between CASI and PAPI. The main effect for CASI compared with PAPI in a generalized estimating equation (GEE) analysis was an OR (95% CI) of 1.04 (0.92-1.17). Variables were also examined individually, including homosexual partnership (adjusted OR 1.26 95%, CI 0.69-2.29), payment for sex (adjusted OR 0.68 95% CI 0.29-1.59), masturbation (adjusted OR 0.89 95% CI 0.66 1.22) and five or more partners in the past 5 years (OR 0.85 95% CI 0.61 -1.19). CONCLUSION: We found no evidence of a consistent effect of CASI on rates of reporting sexual HIV risk behaviours in this sample. CASI resulted in improvement in internal consistency and a reduction in missed questions.