Estimates of dog population sizes are a prerequisite for delivering effective canine rabies control. However, dog population sizes are generally unknown in most rabies-endemic areas. Several approaches have been used to estimate dog populations but without rigorous evaluation. We compare post-vaccination transects, household surveys and school-based surveys to determine which most precisely estimates dog population sizes. These methods were implemented across 28 districts in southeast Tanzania, in conjunction with mass dog vaccinations, covering a range of settings, livelihoods and religious backgrounds. Transects were the most precise method, revealing highly variable patterns of dog ownership, with human: dog ratios ranging from 12.4:1 to 181.3:1 across districts. Both household and school-based surveys generated imprecise and sometimes inaccurate estimates, possible due to low sample size. Transect data were subsequently used to develop a predictive model for estimating dog populations in districts lacking transect data. We predicted a dog population of 2,316,000 (95% CI 1,573,000-3,122,000) in Tanzania and an average human: dog ratio of 20.7:1. Our modelling approach has the potential be applied to predicting dog population size in other districts where mass dog vaccination is carried out, given census and livelihood data. We recommend transects as a rapid and effective method to refine dog population estimates across large geographic areas and guide dog vaccination programs.