Background and methods: Tuberculosis (TB) remains the leading reported cause of mortality among HIV-positive individuals in low- and middle-income countries (LMIC), but disease prevalence and cause-specific mortality data are based on estimates; reliable data are needed to track progress towards targets for reductions in TB and HIV-related deaths. The aim of this work was to estimate disease prevalence and causes of death (CoD) among people dying after enrolment to three large studies of TB and HIV in South Africa. Literature was systematically reviewed for autopsy studies estimating disease prevalence in HIV-positive adults; studies validating verbal autopsy (VA) for HIV and TB deaths; and studies directly estimating CoD in HIV-positive adults in LMIC. Primary data were collected concerning HIV-positive decedents from three parent studies and HIV-negative controls. Minimally-invasive autopsy (MIA), involving tissue biopsy, fluid aspiration, and bronchoalveolar lavage was conducted in a subset of HIV-positive decedents. CoD were assigned, through structured review of clinical and research data (“reference-standard”) and VA data, collected using the World Health Organization (WHO) 2012 instrument and interpreted using physician-certified (PCVA) and computer-coded VA (CCVA) methods (InterVA-4 and SmartVA-Analyze). VA-assigned and reference-standard CoD were compared; agreement was measured at individual- and population-level. Results: MIA was conducted for 34 HIV-positive adults: 16 (47%) had evidence of TB at autopsy, 14/16 (88%) had evidence of extrapulmonary disease, and 6/16 (38%) had not been started on TB treatment; 23/34 (68%) had evidence of bacterial pneumonia and 20/34 (59%) had evidence of two or more concomitant infections. Most (94%) individuals who underwent MIA were assigned HIV-associated reference-standard CoD; this was underestimated by all three VA methods (PCVA 74% [chance-corrected concordance {CCC} 0.71], InterVA-4 47% [CCC 0.42], and SmartVA-Analyze 41% [CCC 0.31]). Reference-standard CoD were assigned, without MIA data, to 259 HIV-positive adults: 183 (71%) were assigned HIV-associated causes. Only the PCVA estimate was similar, at 80% (CCC 0.78); InterVA-4 and SmartVA-Analyze underestimated the HIV-associated mortality fraction (estimates 48% and 29%; CCC 0.48 and 0.20, respectively). An autopsy study exploring the spectrum of disease in individuals with advanced HIV in primary care clinics in South Africa Agreement between VA methods and the reference-standard was poor at individual level (overall CCC ≤0.22) and slightly better at population level (cause-specific mortality fraction [CSMF] accuracy 0.43–0.79). Only PCVA could estimate the HIV-associated TB mortality fraction, underestimating it when compared with a reference standard that included autopsy data (reference 41% vs. PCVA 32%; CCC 0.23) and overestimating it when compared with a reference standard without autopsy data (reference 27% vs. PCVA 42%; CCC 0.42). Among 356 HIV-positive and 103 HIV-negative adults with confirmed HIV status, the VA instrument was sensitive (84.3%) and specific (94.2%) in assigning HIV status; VA methods showed high specificity (all methods >89%) in assigning HIV-associated CoD; and both CCVA methods underestimated the likely true HIV-associated mortality fraction among confirmed HIV-positive decedents (InterVA-4 44.7% and SmartVA-Analyze 22.5%). Conclusions: TB remains a leading CoD among HIV-positive adults in LMIC. Changes are needed to disease classification systems and automated VA methods to allow for better estimation of HIV-associated mortality overall and mortality due to HIV-associated TB. Structured guidelines for assigning CoD in HIV-positive people in clinical settings and the use of MIA at sentinel surveillance sites may be useful additions to current methods.