An ensemble of simulations performed with a coarse resolution 3-D climate model driven by various combinations of external forcing is used to investigate possible causes for differences noticed in two recent simulations of the climate of the past millennium using General Circulation Models (GCMs). Our results strongly suggest that differences in sensitivity (equilibrium and transient climate response) could be responsible for temperature changes that differ by more than a factor of two between two models. In addition, the spin-up procedure could explain some differences between the simulations during the first centuries of the second millennium. The choice of the forcing reconstruction is found to play a smaller role for the differences in the simulated climate, in the model configurations analyzed here. Furthermore, at decadal scale, internal climate variability can mask the differences associated with different forcing reconstructions.