Regime-dependent evaluation is a relatively new approach to assess model performance. It consists of classifying the model biases according to a discrete number of regimes and evaluating model output within each regime. In this paper, the regimes are firstly defined by the large-scale atmospheric circulation, based on the objective Jenkinson-Collison classification technique which distinguishes synoptic patterns by strength, direction and vorticity of the geostrophic flow. Eight directional and two vorticity circulation regimes (circulation types) are specified. In this way, it is possible to quantify the model performance for cases with for example westerly winds only, or with cyclonic circulation only. A second regime classification is based on temperature, which allows for detection of temperature-dependent model performance. Modelled accumulated precipitation (mm/6 h) is evaluated with rain gauges for the years 2007 and 2008. Two variants of the COSMO model are evaluated: a fine-resolution version (2.8 km, COSMO-DE) and a coarse-resolution version (7 km, COSMO-EU). In COSMO-EU, a windward/leeward effect becomes visible since circulation is related to dominant wind direction, hence to windward and lee side of orography. In COSMO-DE, no circulation dependent but a height-related bias is identified and further explored, making use of temperature-dependent evaluation which unveils a positive model bias related to solid precipitation.