The physical factors controlling total mercury (HgT) and methylmercury (MeHg) concentrations in lakes and streams of northeastern USA were assessed in a regional data set containing 693 HgT and 385 corresponding MeHg concentrations in surface waters. Multiple regression models using watershed characteristics and climatic variables explained 38% or less of the variance in HgT and MeHg. Land cover percentages and soil permeability generally provided modest predictive power. Percent wetlands alone explained 19% of the variance in MeHg in streams at low-flow, and it was the only significant (p < 0.02) predictor for MeHg in lakes, albeit explaining only 7% of the variance. When stream discharge was added as a variable it became the dominant predictor for HgT in streams, improving the model r 2 from 0.19 to 0.38. Stream discharge improved the MeHg model more modestly, from r 2 of 0.25 to 0.33. Methylation efficiency (MeHg/HgT) was modeled well (r 2 of 0.78) when a seasonal term was incorporated (sine wave with annual period). Physical models explained 18% of the variance in fish Hg concentrations in 134 lakes and 55% in 20 reservoirs. Our results highlight the important role of seasonality and short-term hydrologic changes to the delivery of Hg to water bodies. ?? 2005 Springer Science+Business Media, Inc.