The power of a validation strategy (that is, its ability to discriminate between good and bad model hypotheses) depends on what kind of data are available and how the data are used to challenge the hypothesis. Several validation strategies are examined from the perspective of power and practical applicability. It is argued that validation using multiresponse data in a catchment experiencing a shift in hydrologic regime due to disturbance or extreme climatic inputs is a considerably more powerful strategy than traditional split-sample testing using streamflow data alone in undisturbed catchments. A case study testing two model hypotheses is presented using paired catchments for which multiple-response data in the form of streamflow, stream chloride, and groundwater levels were available. The first catchment, Salmon, was maintained as an established forest, while the second, Wights, was clear-felled and converted to pasture about 3 years after monitoring started. The hypotheses consider the same lumped hydrosalinity model with the first (H1) excluding a groundwater discharge zone and the second (H2) including it. It was found that even with three concurrent responses from the undisturbed Salmon catchment, H1 could not be rejected, leaving an important part of the model conceptualization unidentified. Moreover, a streamflow split-sample test for the disturbed Wights catchment failed to conclusively reject H1; parameters could be found which accurately tracked the streamflow changes following forest clearing yet produced erroneous simulations of responses such as stream chloride and groundwater storage. It was only when H1 was subjected to the scrutiny of three catchment responses from the disturbed Wights catchment that it could be rejected. This highlights the importance of challenging model hypotheses under the most demanding of tests, which, in this study, coincided with multiple-response validation in a disturbed catchment.¿ 1997 American Geophysical Union