Empirical studies of the interaction between the anther smut fungus Microbotryum violaceum and its host plant Lychnis alpina were combined with modelling approaches to investigate how variation in the spatial distribution of host populations influences disease dynamics and variation in resistance. Patterns of disease incidence and prevalence were surveyed in three contrasting systems of natural L. alpina populations where there is substantial variation in spatial structure, ranging from large continuous populations through to small isolated patches. Disease incidence (fraction of populations where disease was present) was highest in the continuous situation, and lowest in the most isolated populations. The reverse was true for prevalence (fraction of individuals diseased). To better understand the long-term ecological and evolutionary consequences of differences in among population spatial structure, we developed a two-dimensional spatially explicit simulation model in which host-population spacing was modelled by varying the percentage of sites suitable for the host. The general patterns of disease incidence and prevalence generated in the simulations corresponded well with the patterns observed in natural populations of L. alpina and M. violaceum; i.e. the fraction of sites with disease increased while the average disease prevalence in diseased populations decreased when host populations became more connected. One likely explanation for the differences in disease incidence and prevalence seen in natural populations is that the evolution of host resistance varies as a function of the degree of fragmentation. This is supported by simulation results that were qualitatively similar to the survey data when resistance was allowed to vary, but not when hosts were assumed to be uniformly susceptible. In the former, the frequency of resistance increased markedly as host populations became more connected.