Mortality has been studied in 15,577 ankylosing spondylitis (AS) patients diagnosed between 1935 and 1957 in the UK, of whom 14,556 received X-ray treatment. By January 1, 1992 over half of the cohort had died. Among the irradiated patients, cancer mortality was significantly greater than expected from the national rates for England and Wales, with a ratio of observed deaths to expected (relative risk, RR) of 1.30, and significant increases individually for leukaemia, non-Hodgkin's lymphoma, multiple myeloma and cancers of the oesophagus, colon, pancreas, lung, bones, connective and soft tissue, prostate, bladder and kidney. Among the unirradiated patients, cancer mortality was lower than expected from national rates (RR = 0.79). Among irradiated patients, the RRs for leukaemia, lung cancer, and all other neoplasms all decreased significantly with increasing time since first treatment following an initial increase. By 35 years after first treatment, the radiation-related excess for lung cancer had completely disappeared, while for other neoplasms the RR remained significantly raised, although at a lower level than in earlier periods. Most irradiated patients received several courses of treatment within a 5-year period. Based on a 1 in 15 random sample, the mean total body dose received in this period was 2.64 Gy, with the heaviest dose to the vertebrae. A linear dose-response model for all neoplasms except leukaemia gave an excess RR of 0.18 Gy-1 in the period 5-24.9 years after first treatment, which decreased significantly to 0.11 Gy-1 in the period more than 25 years after first treatment. There was no evidence that a linear-quadratic model fitted the data better than a linear model. There were significant dose-response relationships individually for cancers of the lung, oesophagus, colon, pancreas, prostate, bladder and kidney.