Interannual variability of global sea-air CO2 flux is investigated by use of a numerical model for general and biogeochemical circulation with the observed interannual variations in sea surface temperature and wind stress from 1961 to 1998. The global variability is 0.23 GtC yr-1 as the standard deviation, and is governed by the variability of 0.13 GtC yr-1 in the equatorial Pacific through the El Ni¿o-Southern Oscillation (ENSO) cycle. The variability in the Southern Ocean is the second largest, 0.07 GtC yr-1. These features are reproduced in both the experiments using the wind stress from an atmospheric circulation model reanalysis (National Centers for Environmental Prediction (NCEP)) and that from an observational data set (Atlas of Surface Marine Data, 1994). Our results agree with the previous ocean-based estimates on the points that the global variability is much smaller than the atmosphere-based estimates (~1 GtC yr-1) and that the global variability is governed by the equatorial Pacific. The variability in the equatorial Pacific is mostly influenced by the surface dissolved inorganic carbon (DIC) convergence in the wind-driven velocity field; the reduced DIC convergence by the weakened easterly causes the reduced sea-to-air CO2 flux during El Ni¿o. Variability of export production has an opposite effect to that of the surface DIC convergence in controlling the sea-air CO2 flux in the ENSO cycle but is smaller than a half of the variability of the surface DIC convergence. Effect of surface temperature through the CO2 solubility is also opposite but is smaller by an order of magnitude. The sea-air CO2 flux variability influenced by the ENSO is almost the same in both the early period (1961--1979) and the late period (1980--1998). The simulated variability in the equatorial Pacific is smaller, about 70% of the observational estimate. This deficiency is partly due to the underestimated effect of surface salinity variation on pCO2 caused by the climatological salinity boundary condition and partly due to the overestimated export production by the neglect of the effect of micronutrients in our production parameterization.