A five-member ensemble with a coupled atmosphere-sea ice-ocean model is used to examine the effects of natural variability on climate projections for the Arctic. The individual ensemble members are initialized from a 300 years control experiment, each starting from different strengths and phases of the Atlantic Meridional Overturning Circulation. The ensemble members are integrated for 80 years with a 1% per year increase in the atmospheric concentration of CO2. The main findings are that on decadal time scales, multi-model spread of estimated temperature changes in the Arctic may potentially be attributed to internal variability of the climate system. During weak CO2 forcing the internal variability may mask the strength of the anthropogenic signals for several decades. The implications of the findings are that attribution of any Arctic climate change trends calculated over a few decades is difficult.