Distributed hydrological modeling in alpine watersheds needs high-resolution (~1 km in space, 1 h in time) rainfall input. Spatial resolution of operational numerical weather forecast or regional climate models is coarser by at least one order of magnitude. Stochastic downscaling of meteorological model output is the appropriate way to bridge this scale gap. This study examines radar data in an orographically complex area in the southern European Alps by multiplicative multiresolution decomposition. Multiresolution rainfall fluctuations exhibit simple scaling characteristics. This result motivates the implementation of a spatial downscaling model, an example of an anisotropic microcanonical multiplicative random cascade, that is applied and evaluated. Necessity and skill of downscaling is illustrated by hydrological modeling of four heavy precipitation events in an Alpine watershed (total area of 2627 km2) with the grid-based model WaSiM-ETH. Downscaling of rainfall fields with 16 km grid spacing to fields with 1 km grid spacing improves the simulated event-mean runoff by 5--10% and the peak runoff by up to 20%.