Consequences of progressive biodiversity declines depend on the functional roles of individual species and the order in which species are lost. Most studies of the biodiversity–ecosystem functioning relation tackle only the first of these factors. We used observed variation in grassland diversity to design an experimental test of how realistic species losses affect invasion resistance. Because entire plant functional groups disappeared faster than expected by chance, resistance declined dramatically with progressive species losses. Realistic biodiversity losses, even of rare species, can thus affect ecosystem processes far more than indicated by randomized-loss experiments.