To further investigate the relationship between serum total cholesterol (TC) level and mortality due to major causes. In particular, is the elevated mortality among persons with low TC levels due to confounding conditions that both lower TC level and increase the risk of mortality, and is the association between low or high TC level and mortality homogeneous in the population or, alternatively, restricted to persons with other risk factors?Prospective cohort study.Free-living population in Oahu, Hawaii.A total of 7049 middle-aged men of Japanese ancestry.Age- and risk factor-adjusted mortality due to coronary heart disease, hemorrhagic stroke, cancer, chronic obstructive pulmonary disease, nonmalignant liver disease, trauma, miscellaneous and unknown, and all causes.During 23 years of follow-up, a total of 1954 deaths were documented (38% cancer, 25% cardiovascular, and 37% other). Men with low serum TC levels (< 4.66 mmol/L [< 180 mg/dL]) were found to have several adverse health characteristics, including a higher prevalence of current smoking, heavy drinking, and certain gastrointestinal conditions. In an age-adjusted model, and in relation to the reference group (4.66 to 6.19 mmol/L [180 to 239 mg/dL]), those in the lowest TC group (< 4.66 mmol/L [< 180 mg/dL]) were at significantly higher risk of mortality due to hemorrhagic stroke (relative risk [RR], 2.41; 95% confidence interval [Cl], 1.45 to 4.00), cancer (RR, 1.41; 95% Cl, 1.17 to 1.69), and all causes (RR, 1.23; 95% Cl, 1.09 to 1.38). Adjustment for confounders in multivariate analysis (and exclusion of cases with prevalent disease at baseline and deaths through year 5) did not explain the risk of fatal hemorrhagic stroke but reduced the excess risk of cancer mortality by 51% (to 1.20 from 1.41) and reduced the excess risk of all-cause mortality by 56% (to 1.10 from 1.32) in the low TC group. In addition, there were clear differences in the patterns of risk when comparing men with and without selected risk factors (ie, smoking, alcohol consumption, and untreated hypertension).We conclude that the excess mortality at low TC levels can be partially explained by confounding with other determinants of death and by preexisting disease at baseline, and TC-mortality associations are not homogeneous in the population. In our study, TC level was not associated with increased cancer or all-cause mortality in the absence of smoking, high alcohol consumption, and hypertension.