An atmospheric general circulation model (GCM) was forced with the observed near-global sea surface temperature (SST) pattern for the period January 1970-December 1985. Its response over the Pacific Ocean is compared with Tahiti and Darwin station sea-level pressure and wind stress analyses obtained from Florida State University. The time-dependent SST clearly induces in the model run a Southern Oscillation that is apparent in the time series of all considered variables. The phase of the GCM Southern Oscillation is as observed but its low-frequency variance is too low and the spatial pattern is confined mainly to the western Pacific. The model is successful in reproducing the warm events of 1972-73 and 1982-83 and the cold event 1970-71, but fails with the cold events 1973-74 and 1975-76 and with the warm event 1976-77. Because the GCM is used as the atmospheric component in a coupled model, the response of an equatorial oceanic primitive equation model to both the modelled and observed wind stress is examined. The ocean model responds in essentially the same way to forcing with the observed wind stress and to forcing that corresponds to the first two low-frequency empirical orthogonal functions (EOFS) of the wind variations. These first two EOFS describe a regular eastward propagation of the so signal from the western Pacific to the central Pacific within about one year. The ocean model's response to the modelled wind stress is too weak. It is similar to the response to the first observed wind stress EOF only. That is, the observed Southern Oscillation appears as a sequence of propagating patterns but the simulated Southern Oscillation appears as one standing pattern. The nature of the deviation of simulated wind stress from observations is further analysed by means of model output statistics.