Global land surface fluxes of energy and CO2 have been simulated using an off-line version of a biosphere-atmosphere model, SiB2, forced with analyzed or observed atmospheric boundary layer mean potential temperature, water vapor mixing ratio, and wind, surface downward solar and thermal radiation, and precipitation. The off-line model is called SiBDRV. Soil and vegetation boundary conditions were specified from satellite data and other sources. The European Centre for Medium-Range Weather Forecasts (ECMWF) data assimilation system products were used to derive the atmospheric and radiative forcings. Precipitation was based on station observations. The SiBDRV results were compared with corresponding simulation results produced by the Colorado State University general circulation model (CSU GCM), with the ECMWF assimilation system output and with observations. Differences between the surface energy budget components and the surface climatology produced by SiBDRV and the ECMWF assimilation system are due to differences in the land surface parameterizations between the two models. SiBDRV produced lower surface latent heat fluxes and larger sensible heat fluxes than the ECMWF data assimilation, partly due to large canopy resistent term explicitly formulated by SiB2 and possible precipitation differences between the SiBDRV precipitation forcing and the ECMWF data. Differences between the SiBDRV and the CSU GCM results are due to the different climates associated with the ECMWF assimilation system output, which is strongly constrained by assimilated observations, and by the CSU GCM, which is run in pure simulation mode. More specifically, the major reasons for the surface energy and CO2 budget differences between the SiBDRV and the GCM are greater incoming solar radiation in the GCM and differences in the precipitation patterns. The simulated global annual carbon uptake by the terrestrial biosphere is similar in SiBDRV and the GCM. The annual gross primary productions of SiBDRV (116 Gt) and the GCM (113 Gt) agree well with other studies, using either ecological process models or empirical regression models. SiBDRV takes up 10 and 5% more CO2 than the GCM in January and July, respectively. The seasonally varying land surface CO2 fluxes estimated by the SiBDRV and the GCM both compare reasonably well with the results of other calculations. ¿ American Geophysical Union 1996