Biodiversity is declining worldwide from reductions in both species richness and evenness. Field experiments have shown that primary productivity is often reduced when richness of plant species is lowered. However, experiments testing richness effects have used evenness levels that are much higher than normally encountered in plant communities and have been based on the assumption that species extinctions are random. We experimentally varied, for the first time, both species richness (1–8 perennial species/m²) and species evenness (near maximal vs. realistically low) in grassland plots. Net primary productivity (NPP) and ecosystem CO₂ uptake declined when richness was reduced, and reductions were similar between evenness treatments. Richness effects were associated more with a selection effect than with complementarity (found only with high evenness). Importantly, extinctions in plots during the second year were not random, but were greater at low than at high evenness (i.e., with increased rarity) and in species with low aboveground growth rates. Thus, species evenness can affect grassland ecosystem processes indirectly by affecting species richness, and it will be imperative to understand how nonrandom extinctions affect NPP in future studies. Our results indicate that richness studies may not be biased by using mixtures with artificially high evenness levels, but the results also demonstrate that results from these studies are directly applicable only to communities in which plant extinctions are random