The potential for local adaptation between pathogens and their hosts has generated strong theoretical and empirical interest with evidence both for and against local adaptation reported for a range of systems. We use the Linum marginale–Melampsora lini plant-pathogen system and a hierarchical spatial structure to investigate patterns of local adaptation within a metapopulation characterised by epidemic dynamics and frequent extinction of pathogen populations. Based on large sample sizes and comprehensive cross-inoculation trials, our analyses demonstrate strong local adaptation by Melampsora to its host populations, with this effect being greatest at regional scales, as predicted from the broader spatial scales at which M. lini disperses relative to L. marginale. However, there was no consistent trend for more distant pathogen populations to perform more poorly. Our results further show how the coevolutionary interaction between hosts and pathogens can be influenced by local structure such that resistant hosts select for generally virulent pathogens, while susceptible hosts select for more avirulent pathogens. Empirically, local adaptation has generally been tested in two contrasting ways: (1) pathogen performance on sympatric versus allopatric hosts; and (2) sympatric versus allopatric pathogens on a given host population. In situations where no host population is more resistant or susceptible than others when averaged across pathogen populations (and likewise, no pathogen population is more virulent or avirulent than others), results from these tests should generally be congruent. We argue that this is unlikely to be the case in the metapopulation situations that predominate in natural host-pathogen interactions, thus requiring tests that control simultaneously for variation in plant and pathogen populations.