Traditional approaches for the validation of watershed models focus on the goodness of fit between model predictions and observations. It is possible for a watershed model to exhibit a good fit, yet not accurately represent hydrologic processes; hence goodness of fit can be misleading. Instead, we introduce an approach which evaluates the ability of a model to represent the observed covariance structure of the input (climate) and output (streamflow) without ever calibrating the model. An advantage of this approach is that it is not confounded by model error introduced during the calibration process. We illustrate that once a watershed model is calibrated, the unavoidable model error can cloud our ability to validate (or invalidate) the model. We emphasize that model hypothesis testing (validation) should be performed prior to, and independent of, parameter estimation (calibration), contrary to traditional practice in which watershed models are usually validated after calibrating the model. Our approach is tested using two different watershed models at a number of different watersheds in the United States.