To determine more accurately the real-time concentration of fecal indicator bacteria (FIB) in beach water, predictive modeling has been applied in several locations around the Great Lakes to individual or small groups of similar beaches. Using 24 beaches in Door County, Wisconsin, we attempted to expand predictive models to multiple beaches of complex geography. We examined the importance of geographic location and independent variables and the consequential limitations for potential beach or beach group models. An analysis of Escherichia coli populations over 4 yr revealed a geographic gradient to the beaches, with mean E. coli concentrations decreasing with increasing distance from the city of Sturgeon Bay. Beaches grouped strongly by water type (lake, bay, Sturgeon Bay) and proximity to one another, followed by presence of a storm or creek outfall or amount of shoreline enclosure. Predictive models developed for beach groups commonly included wave height and cumulative 48-h rainfall but generally explained little E. coli variation (adj. R² = 0.19-0.36). Generally low concentrations of E. coli at the beaches influenced the effectiveness of model results presumably because of low signal-to-noise ratios and the rarity of elevated concentrations. Our results highlight the importance of the sensitivity of regressors and the need for careful methods evaluation. Despite the attractiveness of predictive models as an alternative beach monitoring approach, it is likely that FIB fluctuations at some beaches defy simple prediction approaches. Regional, multi-beach, and individual beach predictive models should be explored alongside other techniques for improving monitoring reliability at Great Lakes beaches. Copyright ?? 2009 by the American Society of Agronomy, Crop Science Society of America, and Soil Science Society of America. All rights reserved.