Salmonella Dublin infections reduce gross margins and compromise animal health and welfare in dairy cattle herds. Despite on-going control efforts in several countries the duration and risk factors of a persistent infection have been difficult to study due to a lack of suitable data. This study utilised the unique opportunity to extract systematically collected repeated bulk-tank milk antibody measurements from all the Danish dairy herds during a 10-year period to perform a time-to-event analysis of the factors that affect the duration of test-positivity and the hazards of recovery from S. Dublin at herd level. Recovery was defined as a shift from test-positive to test-negative between two year-quarters followed by at least three more test-negative year-quarters. The average duration of infection was approximately 2 years. Predictors of recovery were tested in a multivariable Cox proportional hazard model allowing herds to recover from infection multiple times over the 10-year surveillance period. The model results were based on 36,429 observations with data on all the predictors, representing 3563 herds with a total of 3246 recoveries. Sixty-seven herds (2.4%) remained test-positive throughout the study period. The rest of the 317 herds that did not have any recoveries were censored, mainly due to a cessation of milk production. Prior recovery from test-positivity turned out not to be a significant predictor of recovery in the model. The effect of the duration of infection on the conditional probability of recovery (i.e. the hazard) was time-dependent: early in the study period, long durations of infection were predictive of a low hazard of recovery. Later in the control programme the effect of duration of infection was reduced indicating a desired effect of an intensified control programme. There was an increasing tendency towards longer durations and lower hazard of recovery with: (i) increasing herd sizes, (ii) increasing bulk-tank milk somatic cell counts, (iii) increasing local prevalence within a 5km radius, (iv) organic farming and (v) recent purchase of cattle from test-positive herds. Participation in a voluntary paratuberculosis control programme reduced the duration of infection, and there were indications that recovery from S. Dublin infection was stimulated by a centrally organised and targeted control campaign. This is the first large-scale study that investigated duration of infection and predictors of recovery from S. Dublin in cattle herds over an extended period of time. The results provide useful knowledge for the design of control programmes for S. Dublin.