We explore the extent to which stochastic atmospheric variability was fundamental to development of extreme sea surface temperature anomalies (SSTAs) during the 1997--8 El Ni¿o. The observed western equatorial Pacific westerly zonal stress anomalies (τax), which appeared between Nov. 1996 and May 1997 as a series of episodic bursts, were largely reproducible by an atmospheric general circulation model (AGCM) ensemble forced with observed SST. Retrospective forecasts using a hybrid coupled model (HCM) indicate that coupling only the part of τax linearly related to large-scale tropical Pacific SSTA is insufficient to capture the observed 1997 warming; but, accounting in the HCM for all the τax that was connected to SST, recovers most of the strong SSTA warming. The AGCM-estimated range of stochastic τax forcing induces substantial dispersion in the forecasts, but does not obscure the large-scale warming in most HCM ensemble members.