Pathogens play an important part in shaping the structure and dynamics of natural communities. A shared goal of ecology and epidemiology is to predict when a species is most vulnerable to disease. A leading hypothesis asserts that the impact of disease should increase with host abundance, producing a ‘rare-species advantage’. However, the impact of a pathogen may be decoupled from host abundance, because most pathogens infect more than one species, leading to pathogen spillover onto closely related species. We study how the phylogenetic and ecological structure of the surrounding community can be important predictors of disease pressure in a grassland plant community. We formulate predictive models of pathogen sharing using a global database, then use these models to predict disease pressure at the local scale. We find that we can both explain variation in disease across a community of hosts and predict disease pressure for experimentally introduced novel hosts. Our work has implications for the maintenance of biodiversity, epidemiology, biotic resistance against introduced weeds, and the success of managed plants in agriculture and forestry.