Since approximately 60 years rain rates R are derived from the radar reflectivity Z by the so-called Z-R-relations (Z = aRb) that are applied regardless of the precipitation situation (convective, stratiform). The usage of the Z-R relations is one of the main error sources if rain rates are derived from radar data. An improvement is achieved by the classification of rain events into three clusters. For this purpose several parameters are deviated from the volume data (e.g. radial reflectivity gradient, number of bright-band signatures and the mean height of radar-bins with a minimum reflectivity of 15 dBZ) that enable a discrimination between convective and stratiform events. The classification is performed on the factors derived from a principle component analyses with the k-means algorithm. Neural networks are used to relate ten parameters obtained from the vertical profile of reflectivity derived above the stations recording the precipitation at ground with the rain rate. The root mean squared error could be reduced by combination of classification of rain events and the usage of neural networks up to 25 % compared to the Z-R relation operationally used by the German Weather Service.