Migration, that is the study subjects moving from one residential address to another, is a complication for epidemiological studies where exposures to the agent of interest depend on place of residence [corrected]. In this paper we explore migration in cases from a large British case-control study of childhood cancer and natural background radiation. We find that 44% of cases had not moved house between birth and diagnosis, and about two-thirds were living within 2 km of their residence at birth. The estimated dose at the diagnosis address was strongly correlated with that at the birth address, suggesting that use of just the birth address in this case-control study does not lead to serious bias in risk estimates. We also review other individual-based studies of naturally occurring radiation, with particular emphasis on those from Great Britain. Interview-based case-control and cohort studies can potentially establish full residential histories for study subjects and make direct measurements of radiation levels in the dwellings in question. However, in practice, because of study size and difficulties in obtaining adequate response rates, interview-based studies generally do not use full residential histories, and a substantial proportion of dose estimates often derive from models rather than direct measurements. More seriously, problems of incomplete response may lead to bias, not just to loss of power. Record-based case-control studies, which do not require direct contact with study subjects, avoid such problems, but at the expense of having only model-based exposure estimates that use databases of measurements.