When setting standards for protection against ionising radiation it has been usual to extrapolate from experience with high-dose short-term exposure--studies based on atom bomb survivors and patients exposed to radiation therapeutically. Those who work in the nuclear industry are exposed to low-level predominantly gamma radiation for longer periods, and provide an alternative direct source of information. We have combined mortality data from seven cohort studies on nearly 96,000 nuclear industry workers monitored for external radiation in Canada, UK, and USA to assess directly the carcinogenic effects of protracted low-dose exposure to ionising radiation. The excess relative risk for death from leukaemia, excluding chronic lymphocytic leukaemia, was 2.2 per Sv (90% Cl 0.1 to 5.7). This estimate is intermediate between the linear estimate of 3.7 per Sv and the linear-quadratic estimate (as used in recent leukaemia risk assessments) of 1.4 per Sv derived from Japanese atomic bomb survivors' data. The excess relative risk for death from all cancers, excluding leukaemia, was -0.07 per Sv (90% Cl -0.4 to 0.3). This estimate is consistent with a range of risks varying from negative to nearly twice those estimated from atomic bomb survivors (0.18 per Sv). These are the most precise direct estimates so far made of carcinogenic risk after protracted exposure to low-dose ionising radiation. They provide little evidence that the estimates that form the basis of current radiation protection recommendations are appreciably in error.