BACKGROUND: In rural China, mortality surveillance data may be an alternative to primary data collection in clinical trials; SmartVA (verbal autopsy) is also a potential alternative for endpoint adjudication. The feasibility and validity of both need to be assessed. METHODS: We used mortality data from the first 24 months of the China Salt Substitute and Stroke Study (SSaSS) trial and assessed the agreement between (1) mortality surveillance data and face-to-face visits for fact of death; (2) mortality surveillance data and SSaSS adjudication for causes of death; (3) SmartVA and SSaSS adjudication for causes of death; (4) cause-specific mortality fraction of different methods. Face-to-face visits and SSaSS adjudication were taken as reference methods. The agreement was measured by sensitivity, specificity and positive predictive value (PPV) across different 10th Revision of International Statistical Classification of Diseases chapters. RESULTS: One thousand three hundred and sixty-five deaths were included. Mortality surveillance data had 82% sensitivity for fact of death and 81% sensitivity for causes of death, with substantial variances across different disease types and reasonable quality for circulatory death (91% sensitivity and 94% PPV). The sensitivity of SmartVA for causes of death was 61%, with reasonable quality for deaths of external causes of morbidity (90% sensitivity). The leading causes of death from different sources were the same with some variances in the fractions. CONCLUSION: Using mortality surveillance data for fact of death in clinical trials need to account for under-reporting. A face-to-face visit to all participants at the completion of trials may be warranted. Neither mortality surveillance data nor SmartVA provided valid data source for endpoint events.