A thorough understanding of trends in species diversity along environmental gradients requires knowledge of spatial and temporal patterns of resources, in addition to identification of the mechanisms that generate and maintain resource heterogeneity. Accordingly, patchiness of soil water was examined in 50-m x 50-m replicates of three site types along a producitivity gradient in northern Lower Michigan. All sites were occupied by second-growth forests dominated by bigtooth aspen (Populus grandidentata Michx.) in the overstorey. Soil volumetric water content was measured periodically in 1991 and 1992 at 1-m intervals along orthogonal transects superimposed on 5-m x 5-m grids. Plant cover, soil properties, topography, forest floor and climate were investigated as potential factors regulating soil water. Percent cover of all species in the ground flora was recorded in 1-m x 1-m quadrats. In addition, percent transmittance of photosynthetically-active radiation was measured at 5-m intervals to determine if spatial patterns of above- and below-ground resources were congruent. Spatial data were analysed by geostatistical techniques, and non-parametric tests of linear association used to investigate relationships between variables. Mean soil water content and its standard deviation were positively correlated with site productivity. Variograms of high-productivity sites indicated spatial dependence among points separated by distances of <20 m. Nugget variances accounted for 10 to 60% of the total variance, implying that autocorrelation existed at a scale of <1 m. In sites of low and intermediate productivity, random fluctuation occurred about a mean. Pit and mound topography resulted in the formation of relatively intransient configuration of habitats in high-producitivity sites. Uprooting was caused by clay till in the subsoil that apparently inhibited root growth. In less-productive sites, local relief was low. Species diversity increased with site productivity as a result of elevated richness and equitability. This finding is consistent with predictions based on higher spatial variability and heterogeneity of soil water in sites of high productivity. Although species were non-uniformly distributed along the gradient, there was little indication of habitat partitioning within sites. This result may be attributable to artificial classification of microsites. Incongruities existed between spatial patterns of soil water and light availability. Coexistence of species may thus be facilitated by the fomration of numerous types of microsite.