Ecosystem productivity commonly increases asymptotically with plant species diversity, and determining the mechanisms responsible for this well-known pattern is essential to predict potential changes in ecosystem productivity with ongoing species loss. Previous studies attributed the asymptotic diversity–productivity pattern to plant competition and differential resource use (e.g., niche complementarity). Using an analytical model and a series of experiments, we demonstrate theoretically and empirically that host-specific soil microbes can be major determinants of the diversity–productivity relationship in grasslands. In the presence of soil microbes, plant disease decreased with increasing diversity, and productivity increased nearly 500%, primarily because of the strong effect of density-dependent disease on productivity at low diversity. Correspondingly, disease was higher in plants grown in conspecific-trained soils than heterospecific-trained soils (demonstrating host-specificity), and productivity increased and host-specific disease decreased with increasing community diversity, suggesting that disease was the primary cause of reduced productivity in species-poor treatments. In sterilized, microbe-free soils, the increase in productivity with increasing plant species number was markedly lower than the increase measured in the presence of soil microbes, suggesting that niche complementarity was a weaker determinant of the diversity–productivity relationship. Our results demonstrate that soil microbes play an integral role as determinants of the diversity–productivity relationship.