The hydrological problem to correctly partition rainfall at various scales may require scale-dependent models. This study introduces a new method to identify scale-dependent state space models in a hydrological context. In addition, it investigates relations between model structure, resolution, and uncertainty for the case of overland flow from a hillslope. The main strengths of the method are the ability to evaluate a very large set of candidate models, general applicability, the possibility of interpreting identified models in a physical sense, and computational simplicity. Weaknesses are the possibility complex models that result, computation time required, and the relatively poor fit of models compared to methods with more strict assumptions on errors. For these reasons the method is mainly suitable as a first stage in an identification procedure, leading to the selection of a limited number of candidate models. For the case of overland flow at the hillslope scale it was found that model structure, resolution, and uncertainty are closely related. When compared to the total set of calibrated models, the fittest (best) models are characterized by a small average parameter uncertainty and a constant number of parameters. However, marked structural model differences exist for equally fit models at different spatial and temporal resolutions. Considering the whole spectrum of spatial and temporal resolutions covered by calibrated models, the fittest models are found in a narrow range. ¿ 2000 American Geophysical Union