This paper addresses the problem of why hydrological models often do not give very good results in applications to specific locations with unique characteristics. The implications of uniqueness of place are considered and an approach based on an uncertain or fuzzy mapping of the landscape into a model space is suggested as a way of using field data to condition the predictive uncertainty of models. One consequence of such an approach is the use of hypothesis testing as a way of rejecting some model structures. This allows the semblance of a scientific approach to be restored to modelling activities.