Predicting the likelihood that different plant species will share a common pathogen is central to understanding the role of diseases in maintaining plant diversity, in evaluating the suitability of pathogens as weed biocontrol agents, and for making quarantine decisions about the intentional movement of plants from one region to another.  Most fungal plant pathogens are polyphagous, but the structure of pathogen host ranges remains poorly described.  We cross-inoculated 53 strains of necrotrophic plant pathogenic fungi onto a broad diversity of tropical rain forest plants, and found that the likelihood that a pathogen could infect two plant species decreased continuously with phylogenetic distance between the plants.  This phylogenetic signal in host range continues even to ancient evolutionary distances; congeneric plant pairs had a 67% chance that a pathogen would cause disease on both species, and this probability declined continuously to a probability of 27% at a phylogenetic distance of 285 My.  This quantitative model allows us to predict the likelihood that a pathogen from one host could cause disease on particular plant species, providing an important new tool for disease risk assessment when specific empirical data are not available. Our data suggest that current regulatory approaches strongly underestimate the local risks of global movement of plant pathogens or their hosts.