Abstract Information available 20 years ago suggested that unusual soil conditions were needed to produce high radon concentrations in houses. Based on this belief, simple models were developed to predict house radon concentrations based on estimates of soil radium concentration and soil type (related to permeability). Recent radon-in-housing surveys show that clay soils, which cover much of North America, are radon-prone. The permeability of these soils is not controlled by the soil type, but by the fracture patterns that develop as the soil dries out. Realistic prediction models must take the macroscopic soil structure into account. They must also deal with the effect of the rapid variations in driving force produced by wind velocity variations. The interaction between house and soil is complex, and simple models cannot give realistic results. Many people live in apartment buildings, and are exposed to radon released from building materials. Better models are needed for this source to help develop better controls. Many radon-prone areas have been identified by house surveys, so modelling is now needed only to explain what we know. The priority task today is not prediction, but encouraging production of low-radon dwellings in known high-radon areas.