Data of two years of observations (2007-2008) from the General Observation Period (GOP) are used to evaluate forecasts of the operational COSMO model applications (COSMO-DE and COSMO-EU) of the German Weather Service (DWD). As part of the German Priority Programme on Quantitative Precipitation Forecasting (PQP), the GOP gathered a comprehensive data set from existing instrumentation not used in routine verification and corresponding model output. In this paper we focus on the water cycle variables: integrated water vapor (IWV), cloud base height (CBH) and precipitation. In addition brightness temperatures (BT) from satellite observations are included. The biases in IWV and BT 6.2 μm data are small for COSMODE and COSMO-EU. CBH data show a larger bias with a maximum in the summer season. The largest biases have been found in the precipitation and BT 10.8 μm data. The latter can probably be explained by deficiencies in modelled clouds in the upper troposphere. A classification into different weather condition types gives some additional insight into model deficits. For northerly/north-westerly (maritime) flows model forecasts are too dry (cold) and for southerly (continental) flows too humid (warm).