This paper argues that there are fundamental problems in the application of physically-based models for practical prediction in hydrology. These problems result from limitations of the model equations relative to a heterogeneous reality; the lack of a theory of subgrid scale integration; practical constraints on solution methodologies; and problems of dimensionality in parameter calibration. It is suggested that most current applications of physically-based models use them as lumped conceptual models at the grid scale. Recent papers on physically-based models have misunderstood and misrepresented these limitations. There are practical hydrological problems requiring physically-based predictions, and there will continue to be a need for physically-based models but ideas about their capabilities must change so that future applications attempt to obtain realistic estimates of the uncertainty associated with their predictions, particularly in the case of evaluating future scenarios of the effects of management strategies.