Elevated mercury concentration has been documented in a variety of fish and is a growing concern for human consumption. Here, we explore the influence of physiochemical and watershed attributes on mercury concentration in walleye (Sander vitreus, M.) from natural, glacial lakes in South Dakota. Regression analysis showed that water quality attributes were poor predictors of walleye mercury concentration (R² = 0.57, p = 0.13). In contrast, models based on watershed features (e.g., lake level changes, watershed slope, agricultural land, wetlands) and local habitat features (i.e., substrate composition, maximum lake depth) explained 81% (p = 0.001) and 80% (p = 0.002) of the variation in walleye mercury concentration. Using an information theoretic approach we evaluated hypotheses related to water quality, physical habitat and watershed features. The best model explaining variation in walleye mercury concentration included local habitat features (W(i) = 0.991). These results show that physical habitat and watershed features were better predictors of walleye mercury concentration than water chemistry in glacial lakes of the Northern Great Plains.