Cytomegalovirus (CMV) infects 40 % of the world population and has been suggested to be associated with diabetes; however, no prospective study has ever examined diabetes mortality associated with the infection. A cohort of 14,404 non-diabetic adult participants aged 17-90 years from the Third National Health and Nutrition Examination Survey (1988-1994) was prospectively followed for mortality through 2006. CMV immunoglobulin G was measured by enzyme-linked immunosorbent assay and immunofluorescence assay. Diabetes death was assessed with death records from the National Death Index. Cox proportional hazards modeling was used to determine diabetes mortality risk associated with CMV infection, adjusting for socio-demographics, diabetes risk factors, and comorbidities. At baseline, 76.8 % of subjects were CMV seropositive, and after an average follow-up of 13.7 years, diabetes mortality rate per 10,000 person-years was 6.8 (95 % CI 5.7, 8.0). Among seropositive participants, the diabetes death rate (8.4, 95 % CI 7.0, 9.9) was more than four times the rate in seronegative ones (2.0, 95 % CI 1.1, 3.6) (P value for the difference <0.001). In the adjusted Cox proportional hazards analysis, CMV seropositivity more than doubled the risk of diabetes mortality (HR 2.06, 95 % CI 1.05, 4.06). CMV infection may thus predict future mortality from diabetes in non-diabetic people.