An atmospheric transport model, NIRE‐CTM‐96, was evaluated by using measured radon‐222 concentrations. The model has 2.5×2.5 degree horizontal resolution and 15 vertical levels. Assimilated global meteorological data for 1990–1996 from the European Centre for Medium Range Weather Forecasts were used to drive the model. We used an emanation rate of radon‐222 of 1 atom cm−2 s−1 over mostly ice‐free land. Simulated concentrations were compared with measured concentrations for 22 sites worldwide including 10 stations in China. Simulated annual mean concentrations for Freiburg, Germany, and Socorro, New Mexico, and for four stations in northern China were consistent with the measured concentrations. Simulated daily concentrations for Ogasawara‐Hahajima, Japan, correlated well with the measured concentrations. Simulated upper tropospheric concentrations for Moffet Field, California, demonstrated the cross‐Pacific transport from central Eurasia and India‐Indochina area. Simulated concentrations for two stations in southern China were almost half of the measured concentrations. Mixing layer depth in the model was consistent with other estimates which indicates higher emanation rate there. Simulated concentrations for the South Indian Ocean and the Antarctic during summer were significantly lower than the measured concentrations; this difference was accounted for when emanation from the ocean at a rate of 0.01 atom cm−2 s−1 was included in the model. The model failed to simulate amplitudes of high‐concentration events at Mauna Loa. These high‐concentration events were possibly a result of filament‐like horizontal structure or laminated vertical structure. The vertical as well as horizontal resolution of the model were supposed to be insufficient to reproduce these fine structures.