During a community-based treatment trial of onchocerciasis with ivermectin, verbal autopsies were employed as one method to assess the safety of the drug. The verbal autopsy questionnaire was designed to determine causes of death and mortality differentials in the treated population. During the 8 months of surveillance here reported, 25 individuals died, yet only 9 of these deaths were certified. Seven of the deaths occurred to individuals who had been treated with ivermectin and the majority of the deaths occurred to children under 5, who were excluded from treatment. The verbal autopsy method was evaluated and validated by comparing the verbal autopsy diagnosis of cause of death to death certificate diagnosis, when available. In addition, verbal autopsies were retrospectively performed for all deaths which had occurred at the hospital during the 6 months preceding the start of the study, if these deaths were traceable to households in the surveillance population. We found that in 80% of the adult deaths, the verbal autopsy and death certificate diagnoses of underlying cause of death agreed. The verbal autopsy was less accurate in diagnosing child deaths which we attribute to the design of the verbal autopsy (being to detect potential drug related deaths in adults) and to the delay between death and interview. We conclude that verbal autopsies are an important addition to surveillance systems in remote areas where the absence or inadequacy of health information systems does not allow a thorough follow-up of all subjects in drug studies.