Many studies at the regional scale have found either negative or hump-shaped relationships between productivity and diversity, and some theories propose that these occur because soil resource heterogeneity is either lower or less important in more productive environments. However, there have been few explicit tests of these theories in natural ecosystems. We evaluated the relationship between soil resource heterogeneity and plant richness within a well characterized system of 30 islands in northern Sweden across which soil fertility and productivity declines, and species richness increases, as a consequence of ecosystem retrogression. On each island we created a spatially explicit grid consisting of 49 sampling points in a 9.5 m quadrat, which we used to quantify spatial heterogeneity of five soil variables (NH4+-N, amino N, PO4−-P, microbial biomass, and decomposition), and plant community composition. Using a hierarchical Bayesian approach, we estimated mean semivariograms of each variable for each island size class to compare three components of spatial heterogeneity: total variability, spatial grain, and patchiness. This analysis showed that variability within islands was usually lowest on small islands, where species richness was highest and productivity lowest; however, NH4+-N and amino N had greater patchiness and spatial grain on small islands. We did not detect any significant across-island correlations between whole-plot plant species richness and either whole-plot standard deviation or coefficient of variation of any soil variable. Using partial Mantel tests, we found that mean correlation coefficients between within-plot plant community composition and the soil variables were never significant for any island size class, and did not differ between island size classes. Our findings do not provide any evidence that soil resource heterogeneity controls the productivity–diversity relationship in this system, and suggests other mechanisms are primarily responsible.