As models represent progressively more complicated systems, like integrated groundwater/surface water systems, it becomes increasingly important to (1) use sensitivity and regression methods to understand the often complex and obscure system dynamics, and (2) evaluate hypotheses about system dynamics using alternative models. The methods for accomplishing some aspects of these analyses have not been thoroughly tested and (or) are insufficiently mature to be useful for large, complex, nonlinear models. These concerns are addressed using groundwater/surface water models developed as part of the first stages of an integrated study of the groundwater, surface-water, and ecological aspects of the Maggia valley, Southern Switzerland. This work addresses two aspects of this problem. First, we demonstrate the use of the leave-one-out cross validation technique for model selection. To our knowledge, cross validation was up to now mostly used to compare linear and nonlinear prediction intervals (Christensen and Cooley, 1999). Second, methods for evaluating a large number of alternative models are tested by ranking the models based on three model discrimination criteria (AICc, KIC, and SSWR) and then comparing the rankings against the predictive capabilities of the models. Predictive capabilities are evaluated by cross validation using a set of observations that were not used in model calibration.