Ecosystems behave in unpredictable ways as environments change, making predictions for important drivers like global change difficult. We used a replicated 2 X 2 X 4 factorial ecosystem experiment to quantify the main effects of soil, site, and "community" (4 different mixes of species) on aboveground biomass production. Tree saplings were harvested from experimental mesocosms after 8 years of growth. Statistical interactions among the main effects were significant as were the species to species interactions including both antagonistic and synergistic effects on total mesocosm biomass. In addition, community and species patterns in biomass changed from one abiotic factor combination (e.g., soil X site) to the next, indicating a statistical interaction in species interactions occurred across abiotic factors. We refer to these factor interactions as contextual interactions to distinguish them from species interactions (including competition). These contextual interactions comprise an important component of total variation, and without the benefit of a replicated experimental design, would be misinterpreted as random variation in ecosystem response to main effects such as climate change. Given the difficultly of true replication in large-scale ecosystem experiments, these contextual interactions remain hidden in the guise of biocomplexity.