With increasing model complexity there is a pressing need for new methods that can be used to mine information from large volumes of model results and available data. This work explores strategies to identify and evaluate the causes of discrepancy between models and data related to hydrologic processes, and to increase our knowledge about watershed input-output relationships. In this context, we evaluate the performance of the abcd monthly water balance model for 764 watersheds in the conterminous United States. The work required integration of the Hydro-Climatic Data Network dataset with various kinds of spatial information, and a diagnostic approach to relating model performance with assumptions and characteristics of the basins. The diagnostic process was implemented via classification of watersheds, evaluation of hydrologic signatures and the identification of dominant processes. Knowledge acquired during this process was used to test modifications of the model for hydrologic regions where the performance was "poor".