A retrospective cohort study was conducted to identify risk factors for bovine tuberculosis (bTB) recurrence in New Zealand cattle and deer herds identified as bTB-infected from 1 June 2006 to 1 November 2010. A Cox proportional hazards model identified a positive relationship between the daily hazard of bTB recurrence and: (1) the number of prior bTB episodes for two episodes [hazard ratio (HR) 3·22, 95% confidence interval (CI) 1·21–8·60], and for five episodes (HR 89·5, 95% CI 13·8–580), (2) more than one positive bTB case animal at the index episode (HR 2·25, 95% CI 1·19–4·25) and (3) the presence of cleared test-positives at the final test of the index episode. The proportional hazards assumption was violated for the latter variable so a time-dependent covariate was introduced. Up to 2 years post-clearance, the daily hazard of bTB recurrence was greater in herds with test-positives at the final test (HR 2·59, 95% CI 1·30–5·13), but this effect was not observed more than 2 years' post-clearance (HR 1·05, 95% CI 0·28–3·91). We conclude that unresolved infection contributes to further bTB episodes in the first 2 years after herd clearance.