The Southern Oscillation (SO) is examined in three 10 year AMIP-type integrations of a 30-level GCM having prescribed monthly mean observed sea surface temperatures for the period January 1979 to December 1988. Three horizontal spectral resolutions of T21, T42 and T79 are investigated and the results are compared with the low-frequency variability, having periods longer than 8 months, in the observed Darwin and Tahiti sea level pressures (SLP) and in the T106 ECMWF analyses from May 1985 to April 1991. Both the ECMWF analyses and the GCM results give unrealistic SLP variability at Tahiti resulting in low Darwin-Tahiti SLP correlations and low S/N ratios for the Tahiti-Darwin SO index. The ECMWF analyses are in particularly poor agreement with the observations during 1987 with anomalously high SLP at Tahiti. Examination of the ECMWF assimilated SSTs, reveals that this may be related to the assimilated SSTs being too cold in the central and eastern equatorial Pacific during mid-1987. The GCM results show the familiar SLP dipole in the tropical Pacific albeit displaced eastwards compared to previous observational studies especially at T42 resolution, thus accounting for the problems at Tahiti which lies near strong gradients in the correlation pattern. Time-longitude diagrams of low-level convergence and correlation maps of upper level streamfunction suggest that the model is reproducing the SO divergence anomalies although too weakly at T21 resolution and at different longitudinal locations at T42 and T79 resolutions. The time-mean low-level convergences in the GCM simulations give ITCZs and SPCZs in qualitative agreement with the observations with a tendency for increased convergence in the eastern Pacific ITCZ at higher resolution. Longitudinal shifts are not apparent in the time-mean convergence when comparing the GCM results at different resolutions.