The advent of molecular epidemiology has greatly improved our ability to identify the population sources and track the pathogen movement. Yet the wide spatial and temporal scales usually considered are useful only to infer historical migration pathways. In this study, Bayesian genetic assignments and a landscape epidemiology approach were combined to unravel genetic origin and annual spread during a single epidemic of a plant pathogen: the poplar rust fungus Melampsora larici-populina. The study focused on a particular area—the Durance River valley—which enabled inoculum sources to be identified and channelled spread of the epidemic along a one-dimensional corridor. Spatio-temporal monitoring of disease showed that the epidemic began in the upstream part of the valley and spread out downstream. Using genetic assignment tests, individuals collected at the end of the epidemic were sorted into two genetic groups; very few hybrids were detected, although individuals from both groups coexisted locally downstream in the valley. The epidemic originated from two genetically distinct inoculum sources. Individuals of each group then dispersed southwards along the Durance River and became mixed in poplar riparian stands. These two genetic groups were found previously at a wider spatial scale and proved to result from distinct evolutionary histories on either wild or cultivated poplars. This study showed that the two groups can mix during an epidemic but do not hybridize because they then reproduce asexually. In general, the methods employed here could be useful for elucidating the genetic origin and retracing the colonization history and migration pathways of recent epidemics.