Understanding how an emergent pathogen successfully establishes itself and persists in a previously unaffected population is a crucial problem in disease ecology. In multi-host pathogen systems this problem is particularly difficult, as the importance of each host species to transmission is often poorly characterised, and the epidemiology of the disease is complex. Opportunities to observe and analyse such emergent scenarios are few.Here, we exploit a unique dataset combining densely-collected data on the epidemiological and evolutionary characteristics of an outbreak of Mycobacterium bovis (M. bovis, the causative agent of bovine tuberculosis, bTB) in a population of cattle and badgers in an area considered low-risk for bTB, that has no previous record of either persistent infection in cattle, or of any infection in wildlife.We analyse the outbreak dynamics using a combination of mathematical modelling, machine learning and Bayesian evolutionary analyses. Comparison to M. bovis whole-genome sequences from Northern Ireland confirmed this to be a single introduction of the pathogen from the latter region, with evolutionary analysis supporting an introduction directly into the local cattle population at least six years prior to its first discovery in badgers. Once introduced, the evidence supports M. bovis epidemiological dynamics passing through two phases, the first dominated by cattle-to-cattle transmission before becoming established in the local badger population.These findings emphasise the importance of disease surveillance for early containment of outbreaks, in particular for pathogens not causing immediately evident symptoms in the infected host, and highlight the utility of combining dynamic modelling and phylogenetic analyses for understanding the often complex infection dynamics associated with emergent outbreaks.