Abstract The cold tongue in the tropical Pacific extends too far west in most current ocean–atmosphere coupled GCMs (CGCMs). This bias also exists in the relatively high-resolution SINTEX-F CGCM despite its remarkable performance of simulating ENSO variations. In terms of the importance of air–sea interactions to the climatology formation in the tropical Pacific, several sensitivity experiments with improved coupling physics have been performed in order to reduce the cold-tongue bias in CGCMs. By allowing for momentum transfer of the ocean surface current to the atmosphere [full coupled simulation (FCPL)] or merely reducing the wind stress by taking the surface current into account in the bulk formula [semicoupled simulation (semi-CPL)], the warm-pool/cold-tongue structure in the equatorial Pacific is simulated better than that of the control simulation (CTL) in which the movement of the ocean surface is ignored for wind stress calculation. The reduced surface zonal current and vertical entrainment owing to the reduced easterly wind stress tend to produce a warmer sea surface temperature (SST) in the western equatorial Pacific. Consequently, the dry bias there is much reduced. The warming tendency of the SST in the eastern Pacific, however, is largely suppressed by isopycnal diffusion and meridional advection of colder SST from south of the equator due to enhanced coastal upwelling near Peru. The ENSO signal in the western Pacific and its global teleconnection in the North Pacific are simulated more realistically. The approach as adopted in the FCPL run is able to generate a correct zonal SST slope and efficiently reduce the cold-tongue bias in the equatorial Pacific. The surface easterly wind itself in the FCPL run is weakened, reducing the easterly wind stress further. This is related with a weakened zonal Walker cell in the atmospheric boundary layer over the eastern Pacific and a new global angular momentum balance of the atmosphere associated with reduced westerly wind stress over the southern oceans.