Summary We present here the results of the largest study of childhood cancer and ethnic group in Britain, based on 7,658 children treated at paediatric oncology centres throughout the country. Incidence rates could not be calculated and so relative frequencies were analysed by the log-linear modelling method of Kaldor et al. (1990) with allowance made for regional variations in the ages and diagnostic groups of the children included in the study. Children of Asian (Indian sub-continent) Childhood neoplasms are more appropriately classified by histology rather than primary site. A classification scheme was developed on the basis of histological type by Birch and Marsden (1987) and used for the recent monograph on international childhood cancer incidence (Parkin et al., 1988a). We have used a slightly modified version of this scheme for all of the analyses presented here. In this version, the principal changes are that megakaryocytic leukaemia has been. included with acute non-lymphocytic leukaemia (ANLL), rhabdoid renal tumour and bone-mestastasising renal tumour of childhood have both been included with Wilms' tumour, and peripheral neuroectodermal tumours have been classified with Ewing's sarcoma if in bone and with soft-tissue sarcoma other than rhabdomyosarcoma and fibrosarcoma if in other sites. Langerhans cell histiocytosis has been excluded since it is not now regarded as a neoplasm.The register is not population-based and population data are not available by ethnic group, hence incidence rates could not be calculated and the analyses are based on relative frequencies of different diagnostic groups.The distribution of ethnic groups in Britain varies geographically, and the age distribution of the child population may differ between ethnic groups. The proportion of children referred to paediatric oncology centres, and hence included in the register, varies between regions and diagnostic groups, and older children are generally less likely to be referred (Stiller, 1988). To allow for these confounding factors, the data were analysed by the log-linear modelling method of Kaldor et al. (1990) using the GLIM statistical package (Payne, 1987).As the data were not population-based and incidence rates could not be calculated, a case-control approach was adopted for analysing relative frequencies of each diagnostic group between the ethnic groups (Breslow & Day, 1987