There is currently much interest in understanding how loss of biodiversity might alter ecological processes vital to the functioning of ecosystems. Unfortunately, ecologists have reached little consensus regarding the importance of species diversity to ecosystem functioning because empirical studies have not demonstrated any consistent relationship between the number of species in a system and the rates of ecological processes. We present the results of a simple model that suggests there may be no single, generalizable relationship between species diversity and the productivity of an ecosystem because the relative contributions of species to productivity change with environmental context. The model determined productivity for landscapes varying in species diversity (the number of species in the colonist pool), spatial heterogeneity (the number of habitat types composing the landscape), and disturbance regimes (+/− a non-selective mortality). Linear regressions were used to relate species diversity and productivity for each of the environmental contexts. Disturbance changed the form of the diversity/productivity relationship by reducing the slope (i.e. the change in productivity per species added to the colonist pool), but spatial heterogeneity increased or decreased this slope depending on the particular habitat types composing the landscape. The cause of the diversity/productivity relationship also changed with environmental context. The amount of variation in productivity explained by species diversity always increased with spatial heterogeneity, while the amount of variation explained by species composition (i.e. the particular species composing the colonist pool) tended to increase with disturbance. These results lead us to conclude that the form and cause of the relationship between species diversity and productivity may be highly dynamic-changing over both time and space. Because the trends resulted from well-known mechanisms by which environmental variation alters the absolute and relative abundances of taxa, we suspect this conclusion may be applicable to many different systems.