This paper is addressed towards the problem of extracting appropriate model structures by systematic analysis of rainfall-runoff relationships in gauged catchments. The Upper Enns catchment in the Austrian Alps is selected as the basis of this study. The downward approach championed by Klemes is followed, which involves stepwise adjustment of model structure to capture the observed streamflow variability progressively at the annual, monthly, and then on to daily time scales. Throughout, we focus on emergent properties of the hydrological system at the various time scales, as detected in key signature plots and hydrographs, and model complexity is always kept to the minimum required. Any further alteration or calibration of parameter values is avoided, either with change of scales or in response to inadequate predictions. The downward approach presented leads to parsimonious water balance models with excellent performance and the minimum set of parameters, with a good balance being achieved between model performance and complexity. Copyright (C) 2003 John Wiley Sons, Ltd.