One cornerstone of ecological theory is that nutrient availability limits the number of species that can inhabit a community. However, the relationship between the spatial distribution of limiting nutrients and species diversity is not well established because there is no single scale appropriate for measuring variation in resource distribution. Instead, the correct scale for analyzing resource variation depends on the range of species sizes within the community. To quantify the relationship between nutrient distribution and plant species diversity, we measured NO(3)(-) distribution and plant species diversity in 16 paired, modified Whittaker grassland plots in Serengeti National Park, Tanzania. Semivariograms were used to quantify the spatial structure of NO(3)(-) from scales of 0.4-26 m. Plant species diversity (Shannon-Weiner diversity index; H ') was quantified in 1-m(2) plots, while plant species richness was measured at multiple spatial scales between 1 and 1000 m(2). Small-scale variation in NO(3)(-) (<0.4 m) was positively correlated with 1-m(2) H ', while 1000-m(2) species richness was a log-normal function of average NO(3)(-) patch size. Nine of the 16 grassland plots had a fractal (self-similar across scales) NO(3)(-) spatial distribution; of the nine fractal plots, five were adjacent to plots that had a non-fractal distribution of NO(3)(-). This finding offered the unique opportunity to test predictions of Ritchie and Olff (1999): when the spatial distribution of limiting resources is fractal, communities should display a left-skewed log-size distribution and a log-normal relationship between net primary production and species richness. These predictions were supported by comparisons of plant size distributions and biomass-richness relationships in paired plots, one with a fractal and one with a non-fractal distribution of NO(3)(-). In addition, fractal plots had greater large-scale richness than paired non-fractal plots (1,0-1000 m(2)), but neither species diversity ( H') nor richness was significantly different at small scales (1 m(2)). This result is most likely explained by differences in the scale of resource variation among plots: fractal and non-fractal plots had equivalent NO(3)(-) variation at small scales but differed in NO(3)(-) variation at large scales (as measured by the fractal dimension). We propose that small-scale variation in NO(3)(-) is largely due to the direct effects of plants on soil, while patterns of species richness at large scales is controlled by the patch size and fractal dimension of NO(3)(-) in the landscape. This study provides an important empirical step in understanding the relationship between the spatial distribution of resources and patterns of species diversity across multiple spatial scales.