In nature, plant biomass is not evenly distributed across species, and naturally uncommon species may differ from common species in the probability of loss from the community. Understanding relationships between evenness and productivity is therefore critical to understanding changes in ecosystem functioning as species are lost from communities. We examined data from a large multi-site grassland experiment (BIODEPTH) for relationships between evenness of species composition (proportional abundance of biomass) and total biomass of communities. For plots which started with the same and even species composition, but which diverged in evenness over time, those with lower evenness had a significantly greater biomass. The relationship between evenness and biomass across all plots was also negative. However, for communities where the most common species represented one of the three largest species in monoculture at that site (inclusion of a large dominant species), the relationship was neutral. Path analyses indicated that three paths contributed to this negative relationship. First, higher species richness decreased evenness, but increased biomass (primarily through an increase in maximum plant size). Contrary to predictions, maximum plant size had either no effect on evenness, or a positive effect (in year 3 plots with a large dominant species), thereby reducing this relationship. In year 2, large variation among species in plant size (as measured in monoculture) both decreased evenness and increased biomass, thus increasing the strength of the negative relationship between evenness and biomass. However, the former effect was only found in plots with a large dominant species, the latter only in plots without a large dominant species. When species richness, maximum plant size, and variation in size were accounted for, in year 2 evenness positively affected biomass in plots that included a large dominant species. Our results are consistent with the view that naturally uncommon species may be unaffected by (or even benefit from) the presence of a large naturally common species, and that uncommon plants may have little ability to increase productivity in the absence of such a species. We conclude that the observed negative relationship between evenness and biomass resulted from multiple direct and indirect effects, the relative strength of which depended in part on the presence of large dominant species.