Background: Verbal autopsy, where a close caregiver or relative of someone who recently died reports on the signs, symptoms and circumstances preceding death, is useful for producing population‐based cause of death estimates. However, the performance of verbal autopsy for older adult deaths is poorly understood. Objectives: To evaluate the performance of verbal autopsy in assigning cause of death for adults aged 50+ in a rural area of Malawi. Methods: Cause of death was assigned to each death with a verbal autopsy in the Karonga Health and Demographic Surveillance site between 2002 and 2017 using two methods: (1) Physician review and (2) in silico verbal autopsy (a Bayesian probabilistic model). We calculated uncertainty in cause of death assignment for each method and calculated disagreement in cause of death between methods. Analyses were stratified by age group and sex. Results: A total of 2378 adult deaths were included (1360 aged 50+). Cause of death assignment showed greater uncertainty at older ages in both methods. For example, 59.7% of men aged 80+ were assigned a specific cause of death using physician review, versus 77.5% of men aged 30–49. Population‐level, broad cause of death distributions were similar across methods, but at the individual level there was over 30% disagreement on broad cause of death categories in those aged 50+. Conclusions: Verbal autopsy becomes more uncertain at assigning cause of death at older ages. The inclusion of any reports of medical diagnoses from a doctor and using a two‐stage process of cause of death assignment (with simple cause of deaths assigned using algorithms and more complex cases being reviewed by physicians) could improve cause of death ascertainment using verbal autopsy at older ages.