As high-dimensional nonlinear systems, regional climate models are sensitive to small perturbations of their initial state. This permits such a model, starting from almost identical states, to develop different dynamics which are equally valid solutions under the same given boundary conditions. The range of solutions generated by this internal variability (IV) is examined for the coupled Arctic regional climate model HIRHAM–NAOSIM using three ensembles. Analyzing the variables mean sea level pressure, sea ice extent, and sea ice thickness, annual cycles of IV are found. While boundary conditions significantly affect the interannual dynamics, the choice of the model version has a larger influence on the annual cycle and the magnitude of IV. Considerations of selected cases imply that links of particular IV states to spatial characteristics of the physical fields are detectable sometime but mostly inconsistent. Similarly, effects of the recent atmospheric circulation on IV states of the sea ice variables are diverse and often only weak. An assessment of the relative importance of IV compared with the overall variability shows that the IV is generally dominated by the external forcing but, depending on the season and region, occasionally exceeds the externally forced variability.