Theory predicts that the temporal stability of productivity, measured as the ratio of the mean to the standard deviation of community biomass, increases with species richness and evenness. We used experimental species mixtures of grassland plants to test this hypothesis and identified the mechanisms involved. Additionally, we tested whether biodiversity, productivity and temporal stability were similarly influenced by particular types of species interactions. We found that productivity was less variable among years in plots planted with more species. Temporal stability did not depend on whether the species were planted equally abundant (high evenness) or not (realistically low evenness). Greater richness increased temporal stability by increasing overyielding, asynchrony of species fluctuations and statistical averaging. Species interactions that favoured unproductive species increased both biodiversity and temporal stability. Species interactions that resulted in niche partitioning or facilitation increased both productivity and temporal stability. Thus, species interactions can promote biodiversity and ecosystem services.