We propose a method for constraining a groundwater flow model both by head observations and concentrations of nonconservative solutes such as calcium, using reaction-path modeling. When calibrating flow models in small watershed in northern Wisconsin using head data alone, we encountered problems of nonuniqueness. However, by coupling the flow models with a plagioclase dissolution model, we were able to greatly reduce the number of plausible flow models. First, by using flow modeling and reaction path modeling in parallel, we tested the consistency of residence times predicted by the flow models with solute concentrations predicted by the geochemical models. Mineral dissolution rate parameters were assumed to be spatially uniform; without this condition the geochemistry data would not provide additional constraints to the flow modeling process. For a more comprehensive test of our models, we used reactive-transport modeling to predict the spatial distribution of ions at each site. The models qualitatively reproduced the observed data and our calibrated silicate dissolution rates closely matched those reported in a field study of nearby site. There were also discrepancies between predictions and observations. We attribute these to transient effects and sediment heterogeneities that were not included in the models. While the resulting models are not unique, our approach demonstrates the ability of fairly simple models to explain much of the observed variability in a complex system. ¿ 1998 American Geophysical Union