The association between diabetes and cancer risk remains controversial. Hence, we examined whether midlife diabetes is related to the risk of cancer in late-life, and whether genetic and early-life environmental factors play a role in this association. This study included 25,154 twin individuals born in 1958 or earlier from the Swedish Twin Registry. Information on cancer diagnosis in late life (aged >= 65) during 1998-2014, was derived from the National Patient and Cancer Registries. Diabetes was ascertained based on self- or informant-reported history, patient registry and antidiabetic medication use. Midlife diabetes was defined when diabetes was diagnosed before 65 years. Data were analyzed following two strategies: (i) unmatched case-control analysis for all participants using generalized estimating equation (GEE) models, and (ii) co-twin control analysis for cancer-discordant twin pairs using conditional logistic regression. Overall, 1,766 (7.0%) had midlife diabetes and 5,293 (21.0%) had cancer in late-life. In multiadjusted GEE models, the odds ratios (95% CIs) of diabetes were 10.55 (2.95-37.67) for pharynx cancer, 5.78 (1.72-19.40) for small intestine cancer, 2.37 (1.14-4.91) for liver cancer and 0.48 (0.35-0.67) for prostate cancer. In people with diabetes, diabetes duration was dose-dependently associated with cancer risk. In conditional logistic regression analysis of 176 prostate cancer-discordant twin pairs, the association between midlife diabetes and prostate cancer in later life became stronger. Midlife diabetes increases the risk of pharynx, small intestine and liver cancers, but reduces prostate cancer risk in late life. Genetic and early-life environmental factors may partially contribute to the diabetes-prostate cancer association.