Radon 222 (222Rn) is used as a tracer to probe and intercompare transport, turbulent mixing, and convective mixing in the Laboratoire de M¿t¿orologie Dynamique (LM) and Goddard Institute for Space Studies (GISS) atmospheric general circulation models (GCMs). Formulations for tracer transport and mixing and their control on the global distribution and time variability of 222Rn, as well as parameterizations for the continental surface source flux, are directly implemented into the two GCMs and run ''in-line.'' Tracer formulations are largely inspired by climate variables (heat, moisture, momentum) formulations in the base GCMs. The comparison of model-calculated 222Rn with observations of time (diurnal, seasonal, sporadic), variability, and spatial (horizontal, vertical) distribution shows partial agreement only. Uncertainties of the sources of 222Rn, in particular of the dependence of 222Rn emanation on soil freezing, are substantial, and the significance and reliability of some of the available observations are low. Model intercomparison is not subject to observation limitations, and it clearly indicates that the boundary layer is more homogeneously mixed in the LMD model, whereas deep convection is more efficient at carrying surface-produced quantities to high tropospheric levels in the GISS model. Resolution also makes a large difference. The LMD model has a finer horizontal grid over most of the globe and is almost systematically better than the GISS model at reproducing sharp fluctuations of 222Rn and seasonal cycles. Our results support that 222Rn could provide an unequivocal absolute measure of the GCM's performances if a more comprehensive observational validation was available. ¿ American Geophysical Union 1995