We describe the behaviour of 23 dynamical ocean-atmosphere models, in the context of comparison with observations in a common framework. Fields of tropical sea surface temperature (SST), surface wind stress and upper ocean vertically averaged temperature (VAT) are assessed with regard to annual mean, seasonal cycle, and interannual variability characteristics. Of the participating models, 21 are coupled GCMs, of which 13 use no form of flux adjustment in the tropics. The models vary widely in design, components and purpose: nevertheless several common features are apparent. In most models without flux adjustment, the annual mean equatorial SST in the central Pacific is too cool and the Atlantic zonal SST gradient has the wrong sign. Annual mean wind stress is often too weak in the central Pacific and in the Atlantic, but too strong in the west Pacific. Few models have an upper ocean VAT seasonal cycle like that observed in the equatorial Pacific. Interannual variability is commonly too weak in the models: in particular, wind stress variability is low in the equatorial Pacific. Most models have difficulty in reproducing the observed Pacific 'horseshoe' pattern of negative SST correlations with internnual Nino3 SST anomalies, or the observed Indian-Pacific lag correlations. The results for the fields examined indicate that several substantial model improvements are needed, particularly with regard to surface wind stress.