Summary Thymus cancer epidemiology has been little investigated, but recent clinical studies have suggested an association with the Epstein-Barr virus. We studied thymus cancer incidence 1963-83 and mortalitiy 1959-86 in England and Wales, using data from the National Cancer Register and national mortality files. Mean age-standardised incidence rates of the tumour were 0.72 per million per annum for males and 0.64 for females; mortality rates were about half of this: 0.43 for males and 0.29 for females. There was no significant change in rates over time, nor any consistent pattern of risk by region of residence. Birth cohort analysis of mortality showed in each sex, lowest risk for persons born during the Second World War. The age distribution of the tumour was unusual: a progressive rise in both incidence and mortality rates occurred in each sex at ages up to 60-69, at which there was a striking peak, more marked for males and for incidence data, with a sharp decline thereafter. Immigrants from China and Cyprus had significantly high proportional registration ratios, but based on small numbers.The thymus gland plays an important role in the maturation of immunocompetent T-lymphocyte cells at younger ages. It enlarges until late puberty and then begins a progressive morphological and functional involution (Roitt et al., 1985). Thymus neoplasms are rare, but since the observation in 1901 that they were associated with myasthenia gravis (Weigart, 1901), there has been interest in their study. Their aetiology remains obscure, however, and their epidemiology is unclear because no population-based study has been carried out (Large et al., 1986); most of the available information comes from case-reports or relatively small hospital series (Wilkins et al., 1966;Bernatz et al., 1973;Salyer & Eggleston, 1976;Maggi et al., 1986 (General Register Office, 1959-1970; Office of Population Censuses and Surveys, 1971-1986).Methods of analysis Directly age-standardised incidence and mortality rates were calculated using England and Wales mid-year populations as denominators and the 'World Standard' population as the standard (Smith, 1987). To reduce random variation when examining secular trends, time series data were smoothed by a moving average of span 5 and weights 1,4,6,4,1/16 (Box & Jenkins, 1970). The figure derived from this smoothing necessarily does not contain separate data points for the two first and two last years.Models based on the assumption that the observed number of cancer occurrences or cancer deaths arose from a Poisson distribution were fitted to test the statistical significance of linear trends in incidence and mortality (Breslow & Day, 1987), using the GLIM (Generalised Linear Interactive Modelling) (Baker & Nelder, 1978) computer package.Birth cohort analyses was conducted for mortality but not for incidence, for which insufficient data were available for satisfactory analysis. Standardised cohort mortality ratios (SCMRs) were calculated for each 5-year birth cohort (Case, 1956;Beral, 1974) using,...