Internally generated climate variability on decadal to centennial time scales was studied with a hierarchy of climate models, including global as well as regional models. The analyses of long-term model integrations over 1000 years with global models of different complexity showed their potential to generate interannual to interdecadal variability without external forcing. Furthermore, the capability of the different models to simulate similar atmospheric flow regimes as the real atmosphere was examined with main emphasis on the Northern Hemisphere. These studies revealed similar atmospheric flow regimes as observed (e.g., North Atlantic Oscillation (NAO), Pacific North-America pattern (PNA)), and provided some evidence for multimodal regime behaviour for coupled models. By the use of the regional model, Arctic climate changes associated with large-scale atmospheric circulation changes and their implications for the interpretation of palaeoclimatic data were analyzed in detail. It was shown that a broadly warm or cold Arctic climate is connected with two distinct circulation regimes of the Arctic atmosphere, characterized by different positions and extensions of the mean tropospheric vortex. In conjunction with the phases of the NAO, significant regional climate changes occur over northern Europe, the North Atlantic and Greenland. However, the simulated interannual variability of winter precipitation, for example over Greenland, is very strong and results in a low correlation between the NAO index and the Greenland precipitation. Hence, a reconstruction of the historic NAO index from Greenland ice accumulation rates implies large uncertainties.