The North Atlantic Oscillation (NAO) exhibits variations at interannual to multidecadal time scales and is associated with climate variations over eastern North America, the North Atlantic, Europe, and North Africa. Therefore, it is very important to understand causes of these NAO variations and assess their predictability. It has been hypothesized, based on observations, that sea surface temperature (SST) and sea-ice variations in the North Atlantic Ocean influence the NAO. We describe results of an ensemble of sixteen experiments with an atmospheric general circulation model in which we used observed SST and sea-ice boundary conditions globally during 1949--1993. We show that multiyear NAO and associated climate variations can be simulated reasonably accurately if results from a large number of experiments are averaged. We also show that the ambiguous results of previous NAO modeling studies were strongly influenced by the ensemble size, which was much smaller than that in the present study. The implications of these results for understanding and predictability of the NAO are discussed. ¿ 2000 American Geophysical Union