Predictions of the 1997--1998 El Ni¿o exhibited a wide range of forecast skill that were dependent, in part, on the wind-driven initial conditions for the ocean. In this study the results of a reduced gravity, primitive equation, sigma coordinate ocean general circulation model are compared and contrasted when forced by several different wind products for the 1997--1998 El Ni¿o/La Ni¿a. The different wind products include atmospheric model winds, satellite wind products, and a subjective analysis of ship and in situ winds. The model results are verified against fields of observed sea level anomalies from TOPEX/Poseidon data, sea surface temperature analyses, and subsurface temperature from the Tropical Atmosphere-Ocean buoy array. Depending on which validation data type one chooses, different wind products provide the best forcing fields for simulating the observed signal. In general, the model results forced by satellite winds provide the best simulations of the spatial and temporal signal of the observed sea level. This is due to the accuracy of the meridional gradient of the zonal wind stress component that these products provide. Differences in wind forcing also affect subsurface dynamics and thermodynamics. For example, the wind products with the weakest magnitude best reproduce the sea surface temperature (SST) signal in the eastern Pacific. For these products the mixed layer is shallower, and the thermocline is closer to the surface. For such simulations the subsurface thermocline variability influences the variation in SST more than in reality. The products with the greatest wind magnitude have a strong cold bias of >1.5 ¿C in the eastern Pacific because of increased mixing. The satellite winds along with the analysis winds correctly reproduce the depth of the thermocline and the general subsurface temperature structure. ¿ 2001 American Geophysical Union