BACKGROUND: Little is known about longitudinal associations of exercise with different types of cancer, particularly in Asian populations. The purpose of this research was to estimate the association between the duration of exercise and all-cause and cancer-specific mortality. METHOD: Data were obtained from the Korean Metabolic Syndrome Mortality Study (KMSMS), a prospective cohort study of 303,428 Korean adults aged 20 years or older at baseline between 1994 and 2004 after exclusion of individuals with missing variables on smoking and exercise. Death certificate-linked data until 31 December 2015 were provided by the Korean National Statistical Office. Cox regression models were constructed to evaluate the associations of exercise with cancer mortality after adjusting for potential confounders such as age, alcohol consumption and smoking status. RESULTS: During the follow-up period of 15.3 years (4,638,863 person-years), a total of 16,884 participants died. Both men and women who exercised showed approximately 30% decreased hazards of mortality, compared to those who did no exercise (hazard ratio (HR) 0.70, 95% confidence interval (CI)=0.68-0.73 for men, HR=0.71, CI : 0.67-0.75). A notable observation of this study is the curvilinear associations between the total duration of exercise per week and cancer mortality, with the lowest risk being observed at the low-to-medium levels of exercise; this trend of associations was found for esophagus, liver, lung, and colorectal cancer mortality in men, and all-cause, all-cancer and lung cancer mortality in women. CONCLUSIONS: Individuals who exercised showed considerably lower all-cause and cancer mortality risks compared with those who did no exercise. Policies and clinical trials aimed at promoting minimal or moderate participation in exercise may minimize cancer mortality risk.