The verbal autopsy (VA) is an epidemiological tool that is widely used to ascribe causes of death by interviewing bereaved relatives of children who were not under medical supervision at the time of death. This technique was assessed by comparison with a prospective survey of 303 childhood deaths at a district hospital in Kenya where medically confirmed diagnoses were available. Common causes of death were detected by VA with specificities greater than 80%. Sensitivity of the VA technique was greater than 75% for measles, neonatal tetanus, malnutrition, and trauma-related deaths; however, malaria, anaemia, acute respiratory-tract infection, gastroenteritis, and meningitis were detected with sensitivities of less than 50%. There may have been unwarranted optimism in the ability of VAs to detect some of the major causes of death, such as malaria, in African children. VA used in malaria-specific intervention trials should be interpreted with caution and only in the light of known sensitivities and specificities.