Model-predicted atmospheric concentrations of 222Rn based on two different 222Rn source terms have been compared with observations in the lower troposphere. One simulation used a globally uniform 222Rn source term from ice-free land surfaces of 1 atom cm2 s1; the other assumed a northwards-decreasing source term (linear decrease from 1 atom cm2 s1 at 30°N to 0.2 atom cm2 s1 at 70°N). Zero emissions were assigned to oceans. The northwards-decreasing source term improved predictions at four out of six stations north of 50°N, reducing the mean prediction/observation ratio from 2.8 to 0.87. In the latitudinal band between 30°N and 50°N, the northwards-decreasing source term resulted in systematic under-prediction of atmospheric 222Rn, whereas the uniform source term provided predictions close to observations. Predictions based on the northwards-decreasing source term were significantly (p< 0.01) better than those based on the uniform source term for an averaged vertical 222Rn profile around 44°N, but were not for one around 38°N. The results indicate that a northwards-decreasing source term could be a more realistic representation of actual 222Rn emissions than a uniform 1 atom cm2 s1 source term. However, the decrease in 222Rn source strength with increasing latitude might not begin at 30°N but somewhat further north. This hypothesis should be investigated through model-independent means.