A predictive model (RIVPACS-type) for benthic macroinvertebrates was constructed to assess the biological condition of 1,087 streams sampled throughout the eastern United States from 1993-2003 as part of the U.S. Geological Survey's National Water-Quality Assessment Program. A subset of 338 sites was designated as reference quality, 28 of which were withheld from model calibration and used to independently evaluate model precision and accuracy. The ratio of observed (O) to expected (E) taxa richness was used as a continuous measure of biological condition, and sites with O/E values <0.8 were classified as biologically degraded. Spatiotemporal variability of O/E values was evaluated with repeated annual and within-site samples at reference sites. Values of O/E were regressed on a measure of urbanization in three regions and compared among streams in different land-use settings. The model accurately predicted the expected taxa at validation sites with high precision (SD = 0.11). Within-site spatial variability in O/E values was much larger than annual and among-site variation at reference sites and was likely caused by environmental differences among sampled reaches. Values of O/E were significantly correlated with basin road density in the Boston, Massachusetts (p < 0.001), Birmingham, Alabama (p = 0.002), and Green Bay, Wisconsin (p = 0.034) metropolitan areas, but the strength of the relations varied among regions. Urban streams were more depleted of taxa than streams in other land-use settings, but larger networks of riparian forest appeared to mediate biological degradation. Taxa that occurred less frequently than predicted by the model were those known to be generally intolerant of a variety of anthropogenic stressors. ?? 2007 American Water Resources Association.