Background: From March 1986 through May 1991, we conducted a randomized nutritional intervention trial, the General Population Trial, in Linxian, China, a region with epidemic rates of squamous esophageal and adenomatous gastric cardia cancers. We found that participants who received selenium, β-carotene, and vitamin E had significantly lower cancer mortality rates than those who did not. In the current study, we examined the relationship between selenium levels measured in pretrial (1985) sera from participants and the subsequent risk of developing squamous esophageal, gastric cardia, and gastric non-cardia cancers during the trial. Methods: This study was designed and analyzed in accord with a stratified case–cohort sampling scheme, with the six strata defined by sex and three age categories. We measured serum selenium levels in 590 case subjects with esophageal cancer, 402 with gastric cardia cancers, and 87 with gastric non-cardia cancers as well as in 1062 control subjects. Relative risks (RRs), absolute risks, and population attributable risk for cancers were estimated on the basis of the Cox proportional hazards models. All statistical tests are two-sided. Results: We found highly significant inverse associations of serum selenium levels with the incidence of esophageal (P for trend <10⁻⁴) and gastric cardia (P for trend <10⁻⁶) cancers. The RR and 95% confidence interval (CI) for comparison of highest to lowest quartile of serum selenium was 0.56 (95% CI = 0.44–0.71) for esophageal cancer and 0.47 (95% CI = 0.33–0.65) for gastric cardia cancer. The population proportion of these cancers that is attributable to low selenium levels was 26.4% (95% CI = 14.45–38.36). We found no evidence for a gradient of serum selenium associated with incidence of gastric non-cardia cancer (P for trend = .96), with an RR of 1.07 (95% CI = 0.55–2.08) for the highest to lowest quartile of serum selenium. Conclusions: Our study supports findings from previous prospective studies and randomized trials that variations in selenium levels affect the incidence of certain cancers. In the United States, where intervention trials of selenium are in the planning stages, consideration should be given to including populations at high risk for squamous esophageal and gastric cardia cancers.