A so-called disaggregation technique is used to combine daily rain gauge measurements and hourly radar composites in order to produce a dataset of hourly precipitation in Germany on a grid with a horizontal resolution of 7 km for the years 2001-2004. This state-of-the-art observation-based dataset of precipitation has a high temporal and spatial resolution and will be extended continuously during the upcoming years. Limitations of its quality, which are due to intrinsic problems with observing the highly variable field of precipitation, are discussed and quantified where possible. The dataset offers novel possibilities to investigate the climatology of precipitation and to verify precipitation forecasts from numerical weather prediction models. The frequency of hourly precipitation in Germany above the detection limit of 0.1 mm/h amounts to 10-30 % in winter, with clear maxima in the mountainous regions, and to 6-20 % in summer, when the spatial variability is considerably reduced. The 95th percentile of the frequency distribution is significantly larger in summer than in winter, with local maxima in the mountainous regions in winter, and in the Alpine Foreland and upper Elbe catchment in summer. It is shown that the operational model COSMO-7 with a horizontal resolution of 7 km captures the geographical distribution of the frequency and of the 95th percentile of hourly precipitation in Germany very well. In contrast, the model is not able to realistically simulate the diurnal cycle of precipitation in any region of Germany during summer.