Representative repeated cross-sectional skin testing for bovine tuberculosis (TB) was conducted over a period of three years in a total of 5377 cattle in three randomly selected woredas (districts) in Ethiopia (Meskan, Woldia and Bako-Gazer) that had never previously been tested for TB. Almost all (99 per cent) of the animals included local zebus kept in traditional husbandry systems. The comparative intradermal tuberculin test with two diagnostic thresholds were used to define positive test results, one according to the World Organisation for Animal Health (OIE) recommended cut-off of more than 4 mm, and the other with a cut-off of >2 mm. Data analysis was performed using a logistic regression model with a random effect at the village level. Applying the OIE definition, the overall representative apparent prevalence of bovine TB in skin test-positive local zebus was 0.9 per cent (95 per cent confidence interval [CI] 0.6 to 1.3 per cent). Using a cut-off of more than 2 mm the overall representative prevalence increased to 4 per cent (95 per cent CI 2.4 to 4.8 per cent). Due to the low apparent prevalence, the true prevalence could be calculated only in Meskan (4.5 per cent) and Bako-Gazer (2.4 per cent) for the more than 2 mm cutoff. With the exception of Meskan, prevalence by woreda did not change significantly over the years. Mycobacterium avium reactor animals were found at all study sites, but there were significant geographical variations. Overall, bulls and oxen were more at risk of being positive reactors (odds ratio [OR] 1.6, 95 per cent CI 1.1 to 2.3; OR 2, 95 per cent CI 1.4 to 2.6, respectively), as were animals in good body condition (OR 2, 95 per cent CI 1.5 to 2.9). Similar results were found at woreda level with the exception of Woldia, where none of the analysed variables was significantly associated with a positive test result.