Numerical output from a GFDL global climate model was used to determine whether the present distribution of rawinsonde stations is adequate to deduce the atmospheric structure and its variability in space and time over the globe. Spatial data gaps were found to cause typical rms wind errors averaged over a hemisphere of 2 to 3 m s−1, increasing for the zonal wind component to 5 or 6 m s−1 at jet stream levels. In temperature the spatial data gaps led to rms errors on the order of 0.5 to 1°C in the free atmosphere, in geopotential height between 20 and 30 gpm in, the upper troposphere, and in specific humidity between 1 and 2 g kg−1 near the surface and about 0.3 g kg−1 at 500 mb. Errors due to instrumental deficiencies, unrepresentativeness of the local soudings, deficiencies in the analysis technique and gaps in the time series were found to he less important than those due to the, spatial gaps, even in the Northern Hemisphere. In the Northern Hemisphere, the rawinsonde network was found to be generally adequate to measure large-scale circulation statistics. However, in the Southern Hemisphere the incorporation of additional data sources (rawinsonde, satellite or otherwise) is necessary, especially for defining the fluxes by the mean meridional and stationary eddy circulations.