AbstractThis study presents high‐resolution numerical simulations in order to investigate the small‐scale precipitation variability in a north Alpine region with an extraordinarily high density of operational rain‐gauge stations. Some of the simulations have been conducted with realistic analysis data, considering two heavy‐precipitation cases that occurred in July and August 2005. Both cases were related to cyclonic northeasterly flow and exhibited a small‐scale precipitation distribution differing substantially from the climatological mean. Most notably, a station lying on the southwestern flank of a major mountain massif recorded precipitation amounts far exceeding all surrounding stations in both cases, although its precipitation climatology does not show outstanding values. Despite some quantitative differences, all real‐case simulations exhibit a precipitation pattern consistent with the observed one. They indicate that the small‐scale precipitation variability, including the unusual local maximum, was dominated by the classical seeder–feeder mechanism combined with downstream drift of the precipitation hydrometeors.In addition, a large number of semi‐idealized simulations have been performed, combining realistic topography with synthetic large‐scale flow conditions. Their main purpose is a systematic investigation of the factors controlling the structure and amplitude of the small‐scale precipitation variability. It is found that the ambient wind direction has the largest impact on the small‐scale precipitation patterns. Selecting a 700 hPa wind direction that differs by about 60° from the actual one yields a precipitation pattern that is completely uncorrelated with the observed data. The pattern correlation with the observed data is also degraded when the ambient wind speed is moved away from its real value, but the impact is not as large as for the wind direction. A less systematic impact is found for temperature, or freezing level, but the most realistic precipitation pattern is still obtained when the specified values approximately match those observed in the real cases. Interestingly, when choosing large‐scale flow conditions that approximately match the real ones, the pattern correlation with the observed rain‐gauge data becomes in both cases at least as high as for the best real‐case simulation. Copyright © 2007 Royal Meteorological Society