The deviations of the distribution of Belgian indoor radon data from the log-normal trend are examined. Simulated data are generated to provide a theoretical frame for understanding these deviations. It is shown that the 3-component structure of indoor radon (radon from subsoil, outdoor air and building materials) generates deviations in the low- and high-concentration tails, but this low-C trend can be almost completely compensated by the effect of measurement uncertainties and by possible small errors in background subtraction. The predicted low-C and high-C deviations are well observed in the Belgian data, when considering the global distribution of all data. The agreement with the log-normal model is improved when considering data organised in homogeneous geological groups. As the deviation from log-normality is often due to the low-C tail for which there is no interest, it is proposed to use the log-normal fit limited to the high-C half of the distribution. With this prescription, the vast majority of the geological groups of data are compatible with the log-normal model, the remaining deviations being mostly due to a few outliers, and rarely to a "fat tail". With very few exceptions, the log-normal modelling of the high-concentration part of indoor radon data is expected to give reasonable results, provided that the data are organised in homogeneous geological groups.