OBJECTIVE: To make quantitative predictions about the magnitude of underreporting of human rabies deaths in the United Republic of Tanzania. METHODS: Human rabies deaths were estimated by using a series of probability steps to calculate the likelihood of rabies developing after the bite of a suspected rabid dog, incorporating field data on the incidence of animal bite injuries, the accuracy of rabies recognition, the distribution of bite wounds, and post-exposure treatment. FINDINGS: Predicted human rabies mortality was estimated to be (a) 1499 deaths per year (95% confidence interval 891-2238), equivalent to an annual incidence of 4.9 (2.9-7.2) deaths/100,000, when active surveillance data on bite incidence were used, and (b) 193 deaths per year (32-409), corresponding to an annual incidence of 0.62 (0.1-1.32) deaths/100,000, when national bite statistics were used. The annual mean number of rabies deaths officially recorded for the same period was 10.8 (7.7-14.0). CONCLUSION: In the United Republic of Tanzania, cases of rabies in humans have been greatly underreported. Dog bite injuries are an accessible source of epidemiological data that may be used to estimate the public health burden of rabies and to monitor epidemiological trends in developing countries.