Data on biological assemblage composition may provide clues about which chemical and physical stressors influence communities. My objective was to evaluate how well nutrient enrichment in streams could be predicted using nationally derived optima and tolerance ranks for macroinvertebrates. A large database of macroinvertebrate abundances across a wide range of nutrient concentrations was used to calculate weighted averages (WAs) for 124 genera and 68 families. These WAs were also transformed into ordinal tolerance ranks (range 1-10). The predictive ability of these optima and ranks was evaluated with data from an independent set of 50 streams where temporally intensive sampling had provided a characterization of nutrient loading. Inferred constituent concentrations and mean tolerance ranks were calculated from a single invertebrate sample at each site. Tolerance ranks and optima generally explained 30-40% of the variation in annual loads and time/flow-weighted concentrations. For several constituents, models explained up to 50% of the variation in the ranks of time-weighted concentrations. Although a substantial amount of variation in nutrient levels was predictable, most model estimates were biased. Biological-based inferences generally overpredicted concentrations at low nutrient levels and underpredicted concentrations at high nutrient levels. The implications of these results to the diagnosis of the causes of ecological impairment is discussed.