There is currently a great deal of interest in developing models to predict stream condition based on measures of both anthropogenic and natural disturbance. Our objective was to develop models to predict macroinvertebrate metrics using measures of watershed disturbance in three distinct ecoregions in Oregon and California. We aggregated macroinvertebrate data from various sources in each ecoregion. We used GIS coverages of land use and land cover to calculate a variety of watershed metrics for use as independent variables in the development of multiple linear regression models based on forward stepwise modeling approach. The best fit models from each ecoregion required only two or three disturbance variables to effectively model macroinvertebrate metrics and models explained 41 to 74 percent of the variation in the metrics. In each ecoregion the best model contained some measure of landuse, but often the model was improved by including a natural disturbance variable. Of the macroinvertebrate metrics modeled effectively, the richness of tolerant macroinvertebrates and Ephemeroptera, Plecoptera and Trichoptera richness were shared among all three ecoregions. Developing predictive models can lead to better understanding of causal linkages in stream ecology and may be useful to managers for predicting stream condition at unsampled sites.