A computer-based file of all Veterans Administration (VA) hospitalisation records for the period 1969-1985 was used to identify and follow for cancer development a cohort of 5,161 white males with pernicious anaemia. A total of 34,915 person-years were accrued, with an average length of follow-up of 6.8 years. A total of 481 cancers were diagnosed, slightly higher than the number expected (SIR = 1.2). Significant excesses were observed for cancers of the buccal cavity and pharynx (1.8) and stomach (3.2), and for melanoma (2.1), multiple myeloma (2.1), myeloid leukaemia (3.7) and other and unspecified leukaemia (4.0). Although the excess for stomach cancer was highest in the first year after diagnosis in a VA hospital, risks of 2-fold or greater persisted throughout the study period. The majority of leukaemias occurred in the first year of follow-up, but some excess risk continued beyond this time. The elevated risk of buccal and pharyngeal cancers may relate to heavy alcohol intake among this population, although risks remained high even when the cohort was restricted to patients without an admission for alcoholism. Although an elevated risk of stomach cancer among pernicious anaemia patients is consistent with most previous surveys, the low absolute risk suggests that the cost-effectiveness of intensive screening should be reassessed.