OBJECTIVE: Verbal autopsies (VAs) are interviews with a relative or friend of the deceased; VAs are a technique used in surveillance sites in many countries with incomplete death certification. The goal of this study was to assess the accuracy and validity of data on HIV status and antiretroviral therapy (ART) usage reported in VAs and their influence on physician attribution of cause of death. DESIGN: This was a prospective cohort study. METHODS: The Karonga Health and Demographic Surveillance Site monitors demographic events in a population in a rural area of northern Malawi; a VA is attempted on all deaths reported. VAs are reviewed by clinicians, who, with additional HIV test information collected pre-mortem, assign a cause of death. We linked HIV/ART information reported by respondents during adult VAs to database information on HIV testing and ART use and analysed agreement using chi-square and kappa statistics. We used multivariable logistic regression to analyse factors associated with agreement. RESULTS: From 2003 to 2014, out of a total of 1,952 VAs, 80% of respondents reported the HIV status of the deceased. In 2013-2014, this figure was 99%. Of those with an HIV status known to the study, there was 89% agreement on HIV status between the VA and pre-mortem data, higher for HIV-negative people (92%) than HIV-positive people (83%). There was 84% agreement on whether the deceased had started ART, and 72% of ART initiation dates matched within 1 year. CONCLUSIONS: In this population, HIV/ART information was often disclosed during a VA and matched well with other data sources. Reported HIV/ART status appears to be a reliable source of information to help classification of cause of death.