BACKGROUND: Audio computer-assisted self-interviews (ACASI) are increasingly used in health research to improve the accuracy of data on sensitive behaviors. However, evidence is limited on its use among low-income populations in countries like India and for measurement of sensitive issues such as domestic violence. METHOD: We compared reports of domestic violence and three less sensitive behaviors related to household decision making and spousal communication in ACASI and face-to-face interviews (FTFI) among 464 young married women enrolled in a longitudinal study of gender-based power and adverse health outcomes in low-income communities in Bangalore, India. We used a test-retest design. At the 12-month study visit, we elicited responses from each participant through FTFI first, followed by ACASI. At the 24-month visit, we reversed the order, implementing ACASI first, followed by FTFI. Univariable log-linear regression models and kappa statistics were used to examine ACASI's effects on self-reports. RESULTS: Regression results showed significantly lower reporting in ACASI relative to FTFI at both visits, including for domestic violence (12-month risk ratio [RR] = 0.61, 95% CI = 0.52, 0.73; 24-month RR = 0.74, 95% CI = 0.62, 0.89). Response agreement between interview modes, calculated by kappa scores, was universally low, though highest for domestic violence (12-month κ = 0.45; 24-month κ = 0.48). Older age and greater educational attainment appeared associated with higher response agreement. CONCLUSIONS: Greater reporting in FTFI may be due to social desirability bias for the less sensitive questions and perceptions of therapeutic benefit for domestic violence. These results cast doubt on the appropriateness of using ACASI for measurement of sensitive behaviors in India.