Wind speeds U obtained using three different sensor types are compared to assess which might be most useful for calculating gas fluxes across the air-sea interface. U remotely sensed using scatterometer (ERS 2 and NASA scatterometer (NSCAT)) and microwave radiometer (special sensor microwave imager (SSM/I)) are compared with each other and with U measured in situ (Carbon Interface Ocean Atmosphere (CARIOCA) drifting buoy). The effect of different gridding schemes is also evaluated for the satellite data. Global comparisons for January 1997 indicate that 1¿ and monthly ERS 2 U are systematically 0.3 to 2 m s-1 lower than NSCAT U at mid and low latitudes. The NSCAT-SSM/I comparisons are similar for the three SSM/Is. SSM/I U appear to be higher by 1 to 2 m s-1 than NSCAT U in the tropics and to be lower at high latitudes. When individual satellite measurements are collocated, ERS 2 U appear to be more coherent with NSCAT (rms of the difference equal to 0.8 m s-1) than SSM/I U (rms of the difference equal to 1.3 m s-1). However, when comparing 1¿ one-week interpolated fields, NSCAT-ERS 2 differences are more scattered than NSCAT-SSM/I differences, because of the high ERS 2 undersampling. This flaw is attenuated when monthly interpolated 1¿ fields are compared. The CO2 exchange coefficients K deduced from these remotely sensed U are compared. Longitudinal profiles of monthly K are consistent within 0 to about 25% depending on the regions and on the instruments. Comparisons with in situ measurements in a region of very low U in the eastern equatorial Pacific show excellent agreement between NSCAT and buoy U (bias of 0.3 m s-1) whereas ERS 2 U are underestimated by 1.3 m s-1 and SSM/I U are overestimated by 0.7 m s-1 on average. ¿ 1999 American Geophysical Union