Childhood cancer is rare everywhere in the world, with age-standardized annual incidence usually between 70 and 160 per million at age 0-14 years. Greater variation is seen between populations for some specific tumour types. Some of the largest variations are geographical and are attributable to environmental factors, whereas variation mainly on ethnic lines seems likely to be a marker of genetic predisposition. A wide range of familial and genetic syndromes is associated with an increased risk of childhood cancer. Virtually all the excess risk of cancer among first-degree relatives of children with cancer can be accounted for by known hereditary syndromes. Studies of weak predisposition and gene-environment interaction have so far shown limited consistency. There are very few established environmental or exogenous risk factors and most of these are infective agents. Many putative risk factors can hardly ever be investigated epidemiologically except by interview or questionnaire studies. Some recent examples illustrate the continuing problems of participation bias and recall bias.