A new version of the ECHAM model is investigated in respect of the hydrological cycle in the Arctic and North European area. Several horizontal and two vertical resolution versions are studied. The higher-resolution ECHAM5 models are, in many respects, superior to the lower-resolution versions of the same model family and the older ECHAM4 model. The vertical resolution has a decisive impact but also increased horizontal resolution leads mostly to improvements. Here T106 (about 110 km) often gives the best results. The summer maxima of precipitation, surface temperature and latent heat flux are simulated too early by about a month for several river catchment areas. This shift is strongest in the T106 and T159 models. Another problem with the annual cycle of precipitation is a relative minimum in August to October, especially in the low-resolution ECHAM5 models. The precipitation of the ECHAM5 simulations over the Arctic region exceeds all observational estimates by 5–15 mm/month, strongest in May–June. The latent heat flux over the river catchments has a clear trend towards increased fluxes with higher horizontal and vertical resolution, which seems to reach a maximum with T106. In the comparison of annual mean P-E (precipitation minus evaporation) with observed river discharge only the horizontal resolution seems to be important, again giving best results for the high-resolution models. The year-by-year variability of the simulations is too high, which is more pronounced for the higher-resolution versions. Especially strong impacts are found from the vertical resolution. The interannual variability of the latent heat flux is much smaller than that of precipitation and therefore the results shown for precipitation apply also for the simulated river discharge. Some forcing of ocean temperature anomalies on the precipitation over the Rhine, Kolyma and Indigirka catchment areas have been found, from the northeastern Atlantic and from the Pacific with developing El Niños. Despite the increased random variability in the higher-resolution models, the signal could be detected in almost all simulations. On the whole the higher-resolution (horizontal and vertical) ECHAM5 model simulations are quite improved compared to the low-resolution version of the same model and an older T42 model version. Increasing the vertical resolution from 19 to 31 levels is decisive for this better performance.