Debates over the relationship between biodiversity and disease dynamics underscore the need for a more mechanistic understanding of how changes in host community composition influence parasite transmission. Focusing on interactions between larval amphibians and trematode parasites, we experimentally contrasted the effects of host richness and species composition to identify the individual and joint contributions of both parameters on the infection levels of three trematode species. By combining experimental approaches with field surveys from 147 ponds, we further evaluated how richness effects differed between randomized and realistic patterns of species loss (i.e. community disassembly). Our results indicated that community-level changes in infection levels were owing to host species composition, rather than richness. However, when composition patterns mirrored empirical observations along a natural assembly gradient, each added host species reduced infection success by 12–55%. No such effects occurred when assemblages were randomized. Mechanistically, these patterns were due to non-random host species assembly/disassembly: while highly competent species predominated in low diversity systems, less susceptible hosts became progressively more common as richness increased. These findings highlight the potential for combining information on host traits and assembly patterns to forecast diversity-mediated changes in multi-host disease systems.