Differential participation between cases and controls can lead to biased estimates of risk. However, the effects of participation are often ignored. We report a detailed analysis of locations of residence for participants and non-participants in a large, national case-control study of childhood cancer in Great Britain, using the 1991 census. The initial selection of 7669 controls, taken from lists of those registered with a General Practitioner, was representative of the British population in respect to an areal-based index of material deprivation. However, parents of controls agreeing to participate were living in more affluent areas than initially selected controls and their matched 3838 cases. The three components of the deprivation index, persons unemployed, households not owning a car or their home were similarly associated with participation. Other census characteristics, such as proportion of flat dwellers and centrally heated households were also associated with control participation. Population density of the local area was not different between participating controls and their matched cases. However, initially selected controls lived in more urban areas than their cases. Such differences are not unique to this study, as they are an inevitable consequence of incomplete participation. The implications of these differences are discussed, in relation to the difficulty this imposes in the interpretation of studies of disease aetiology.