Benthic diatom communities were sampled from natural substrates along a depth transect in northern Lake Michigan to examine spatial and seasonal distribution patterns and define assemblages of associated taxa. Information-theory diversity, evenness, and richness were significantly higher in deep assemblages than in shallow assemblages, with the boundary falling between 5 and 10 m in winter and fall, 10 and 20 m in spring, and 20 and 31 m in summer. Planktonic taxa are abundant on benthic substrates in the deep zone, primarily as dead fallout. In shallow water, large numbers of living planktonic taxa overwinter beneath the ice on benthic substrates. Cluster analysis of samples was used to identify and define diatom assemblages from deep, mid-depth, and shallow water. The deep-water assemblage contains motile and nonmotile benthic species that have been reported from alpine and boreal oligotrophic lakes. The mid-depth assemblage is rich in epipsammic taxa. The shallow-water assemblages contain cosmopolitan taxa that are collected often in nearshore plankton samples. These shallow assemblages do not persist throughout the year as do the deep and mid-depth assemblages. Cluster analysis of taxa provides further definition of species associations and the degree of correlation among species. Correlations among benthic taxa are strengthened when planktonic taxa are removed from the analysis, indicating that plankton fallout represents an addition of information which is poorly related to distribution of the benthic species. The most stable and diverse community is found at the deepest sampling depths where environmental conditions are most stable, the mid-depth community exists in fluctuating but predictable conditions, and the shallow assemblages occur in the zone of maximum physical fluctuation or natural stress. Substrate types are correlated with the species assemblages, but the high diversity of substrate types in shallow water does not support the most diverse assemblage. The discrete species assemblages seen in an oligotrophic area of this large dimictic lake apparently respond sensitively to temporally and spatially dynamic environmental parameters.