Whether and how microparasites such as rabies persist in their host populations are among the fundamental questions of infectious disease epidemiology. Rabies is a fatal disease of all mammalian species, but not all mammalian species can maintain the infection as reservoirs. The approach to control depends on which of the affected species do act as reservoirs. Bringing together old and new data, we examine here the role of wild and domestic animals in maintaining rabies in the Serengeti region of Tanzania, presenting our findings in two parts. In Part I, we argue that domestic dogs are the likely reservoirs because: (1) rabies has been continuously present in the dog population since its (re)introduction in 1977, whilst (2) wildlife cases have been very rare over this period, despite intensive study of Serengeti carnivores; (3) outbreaks of rabies in wild canids (jackals) elsewhere in Africa (Zimbabwe) have followed, rather than preceded, outbreaks in the dog population; (4) all viruses isolated from wild carnivores in the Serengeti ecosystem (including the Kenyan Masai Mara) are antigenically and genetically indistinguishable from the typical domestic dog Strain; (5) dog rabies control in the Serengeti between 1958–77 apparently eliminated the disease from both dogs and wildlife. Having identified dogs as reservoirs, Part II explores some possible mechanisms of maintenance in dog populations. In theory, infection is more likely to be maintained at higher dog densities, and we provide evidence that rabies is maintained in one district with a dog density > 5/km2, but not in two other districts with densities < 1/km2. Because 5 dogs/km2 is much lower than the expected density required for persistence, we go on to investigate the role of atypical infections, showing: (1) from serology, that a substantial proportion of healthy dogs in the Serengeti have detectable serum levels of rabies-specific antibody; (2) from mathematical models that, whilst we cannot be sure what seropositivity means, persistence in low-density dog populations is more likely if seropositives are infectious carriers, rather than slow-incubators or immunes.