A long-term study of disease dynamics caused by the rust Uromyces valerianae in 31 discrete populations of Valeriana salina provides a rare opportunity to explore extended temporal patterns in the epidemiology of a natural host-pathogen metapopulation. Over a 28-year period, pathogen population dynamics varied across the metapopulation with disease incidence (presence/absence), prevalence (% plants infected) and severity (% leaf area covered by lesions) all showing strong population and year effects, indicative of heterogeneity among years and host populations in the suitability of conditions for the pathogen. Disease incidence within individual host populations was significantly affected by host population size, disease prevalence the previous year and the proximity of neighbouring populations infected in the current year. After accounting for these variables there was still a marked temporal component with winter sea level having a significant effect; as did summer rainfall in the second part of the study period (1997-2011). Disease prevalence was also effected by host population size and disease prevalence in the previous year. However, it was less affected by spatial aspects of disease spread than was disease incidence. Winter sea level and June rainfall significantly affected disease prevalence. Assessment of disease impact on plant performance found strong variation in disease severity associated with the aspect and positioning of host populations. Plants growing in lower disease environments produced significantly more seeds than those growing in high disease sites. Significant variation in reaction to infection by U. valerianae was detected among plants within four populations and between these different populations. Synthesis. The epidemiology of U. valerianae was highly influenced by host population size, previous disease and distance. After accounting for these factors, there was a clear temporal signal of change in disease incidence linked to winter sea level and summer rainfall. These patterns reinforce the importance of considering interactions in multiple populations over long periods of time in order to obtain a clear picture of the variability of disease-induced selection pressures across time and space. The behaviour of the pathogen fitted that predicted for a metapopulation with considerable asynchrony in epidemiological patterns among demes.