Increasing time trends in the recorded incidence of childhood cancer have been reported in many different settings. The extent to which these trends reflect real changes in incidence, rather than improvements in methods for diagnosis and registration, is controversial. Using data from the National Registry of Childhood Tumours (NRCT), this thesis investigates time trends in cancer diagnosed under age 15 in residents of Britain during 1966-2005 (54650 cases), and considers potential sources of artefact in detail. Several different methods are used to estimate completeness of NRCT registration. The history of methods for diagnosis and registration of childhood cancers in Britain is described, and predictions are made for effects on recorded incidence. For each of the 12 main diagnostic groups, Poisson regression is used to fit continuous time trends and ‘step’ models to the annual age-sex-standardised rates by year of birth and year of diagnosis. Age-specific rates by period, and quinquennial standardised rates for diagnostic subgroups, are shown graphically. For three broad groups (leukaemia, CNS tumours and other cancer), geographical variation is compared by period of diagnosis. The results of these analyses are discussed in relation to the predicted artefacts. The evidence for a positive association between affluence and recorded incidence of childhood leukaemia is briefly reviewed. A special form of diagnostic artefact, the ‘fatal infection’ hypothesis, is proposed as an explanation of both this association and the leukaemia time trend. This hypothesis is examined in a novel test based on clinical data. The recorded incidence of childhood cancer in Britain increased in each of 12 diagnostic groups during 1966-2005 (from 0.5% per year for bone cancer to 2.5% for hepatic cancer, with 0.7% for leukaemia). Evidence presented here suggests that these increases are probably artefacts of diagnosis and registration. The potential implications for epidemiological studies of childhood cancer should be considered.