Urbanization effect models were developed to differentiate between effects on aquatic macroinvertebrates within a watershed from non-point source urbanization and known local contaminated sediments. Using U.S. Environmental Protection Agency (US EPA) Environmental Monitoring and Assessment Program (EMAP) data from the New England Wadeable Stream Survey (NEWS) and datasets from States of Maine (ME) and Connecticut (CT), we derived macroinvertebrate community response curves for watersheds with different levels of urban development (n = 731). We applied boosted regression trees (BRT) to develop models, allowing us to simultaneously differentiate interactions among variables and quantitatively identify biological effect thresholds with known confidence intervals. Best predictors of watershed development impacts were percent Impervious Area (%IA) at the watershed- or local- scale and percent high density residential area (i.e., with 80 - 100% impervious cover) in the stream buffer. When these indicators operated at both watershed and local scales, they tended to have synergistic (more than additive) effects. For the first time, we were able to demonstrate the effects of road density and road-stream crossings independent of impervious area effects. We also demonstrated declines in community metrics at very low levels of urbanization (<1 - 2% IA), once effects of moderating variables had been factored out. Percent forested buffer was a significant moderating influence on impacts, with sensitivity modified by watershed area, slope class, Ecological Unit (Maxwell et al. 1995), and low flow class. BRTs were powerful enough to discriminate local impacts (Superfund contaminated sediment sites) from upstream development with 95% confidence, once toxic stressor-specific indicators were incorporated.