A major challenge in evaluating patterns of species richness and productivity involves acquiring data to examine these relationships empirically across a range of ecologically significant spatial scales. In this paper, we use data from herb-dominated plant communities at six Long-Term Ecological Research (LTER) sites to examine how the relationship between plant species density and above-ground net primary productivity (ANPP) differs when the spatial scale of analysis is changed. We quantified this relationship at different spatial scales in which we varied the focus and extent of analysis: (1) among fields within communities, (2) among fields within biomes or biogeographic regions, and (3) among communities within biomes or biogeographic regions. We used species density (D=number of species per m2) as our measure of diversity to have a comparable index across all sites and scales. Although we expected unimodal relationships at all spatial scales, we found that spatial scale influenced the form of the relationship. At the scale of fields within different grassland communities, we detected a significant relationship at only one site (Minnesota old-fields), and it was negative linear. When we expanded the extent of analyses to biogeographic regions (grasslands or North America), we found significant unimodal relationships in both cases. However, when we combined data to examine patterns among community types within different biogeographic regions (grassland, alpine tundra, arctic tundra, or North America), we did not detect significant relationships between species density and ANPP for any region. The results of our analyses demonstrate that the spatial scale of analysis – how data are aggregated and patterns examined – can influence the form of the relationship between species density and productivity. It also demonstrates the need for data sets from a broad spectrum of sites sampled over a range of scales for examining challenging and controversial ecological hypotheses.