Non-metric multidimensional scaling (MDS) ordinations were conducted at 5 main taxonomic levels and 1 mixed level (corresponding to major faunal groups) for community data from kelp holdfasts sampled across a gradient of domestic effluent at Coffs Harbour, New South Wales, Australia. All plots showed separation of close outfall sites from controls. Two-way nested analysis and pairwise comparisons of locations using non-parametric randomisation tests (ANOSIM) indicated that differences between the close outfall location and controls were significant at all taxonomic levels. In contrast, the effect due to sites within locations, although significant up to order, was not significant for analyses at the mixed, class and phylum levels. These observations support the hypothesis that anthropogenic effects modify community structure at higher taxonomic levels than natural environmental variables. Consequently, studies of impact that focus on higher taxa may be more readily interpretable than those conducted at the species and family levels.