The present study estimated excess relative risk per sievert (ERR/Sv) of cancer mortality among the cohort of 200 583 male Japanese nuclear workers, with an average individual cumulative dose of 12.2 mSv (<10 mSv, 75.4%; 100 + mSv, 2.6%), conducting Poisson regression using dose category specific observed and expected numbers of deaths, and average doses obtained from the official report of the Radiation Effects Association (REA) on the analysis of mortality of Japanese nuclear industry workers for 1991-2002, which reported the estimates of ERR/Sv for leukaemia but not for all cancers or any other cancer site. The possible confounding biases from drinking alcohol and smoking tobacco were evaluated by examining the association of cumulative radiation dose with the mortality of cancers related to drinking or smoking. For leukaemia (80 deaths), the estimate of ERR/Sv was - 1.93 (95% confidence interval (CI) = - 6.12, 8.57). For all cancers excluding leukaemia (2636 deaths), while the ERR/Sv was estimated to be 1.26 (95%CI = - 0.27, 3.00), confounding by alcohol consumption was suspected since the ERR/Sv estimate of alcohol-related cancers was 4.64 (95%CI = 1.13, 8.91) and the ERR/Sv estimate of all cancers excluding leukaemia and alcohol-related cancers was 0.20 (95%CI = - 1.42, 2.09). In conclusion, confounding by important lifestyle factors related to cancer risk may have a substantial effect on risk estimates, especially when conducting studies of low cumulative dose and, accordingly, low statistical power. Pooled analysis or meta-analysis of nuclear workers for solid cancers needs to take this point into account.