Sleep duration could affect glucose tolerance and mortality. However, the impact that sleep duration has on prognosis of people with diabetes is unclear. A cohort of Japanese self-reported diabetic people (1674 men and 1240 women) aged 40 to 79 years without a history of cardiovascular disease, cancer, renal diseases or pulmonary tuberculosis at baseline (1988 to 1990) were followed until 2009. The average sleep duration was grouped into the following five categories: ≤5 (≤5.4), 6 (5.5-6.4), 7 (6.5-7.4, reference), 8 (7.5-8.4), and ≥9 (≥8.5) hours. Multivariable hazard ratios were calculated for total and cause-specific mortality in each category by using Cox proportional hazard model. During 45 329 person-years of follow-up, 1315 deaths from all-cause (785 men and 530 women), 389 from cardiovascular disease (210 men and 179 women), 359 from cancer (244 men and 115 women), and 138 from infectious disease (87 men and 51 women) were documented. Multivariable hazard ratios (95% CI) were as follows; for men, 1.49 (1.02-2.17) for all-cause in ≤5 h sleepers; 1.53 (1.22-1.91) for all-cause in ≥9 h sleepers; 1.56 (1.01-2.41) for cardiovascular disease in ≥9 h sleepers; 3.08 (1.17-8.12) for infectious disease in ≤5 h sleepers; 2.23 (1.13-4.39) for infectious disease in ≥9 h sleepers; for women, 1.44 (1.09-1.90) for all-cause in ≥9 h sleepers; 2.03 (1.05-3.94) for cancer in ≤5 h sleepers. In conclusion, diabetic people with ≤5 h or ≥9 h of sleep had significantly higher risks of mortality than 7-hour diabetic sleepers.