Twin studies of diabetes mellitus can help elucidate genetic and environmental factors in etiology and can provide valuable biological samples for testing functional hypotheses, for example using expression and methylation studies of discordant pairs. We searched the volunteer Australian Twin Registry (19,387 pairs) for twins with diabetes using disease checklists from nine different surveys conducted from 1980–2000. After follow-up questionnaires to the twins and their doctors to confirm diagnoses, we eventually identified 46 pairs where one or both had type 1 diabetes (T1D), 113 pairs with type 2 diabetes (T2D), 41 female pairs with gestational diabetes (GD), 5 pairs with impaired glucose tolerance (IGT) and one pair with MODY. Heritabilities of T1D, T2D and GD were all high, but our samples did not have the power to detect effects of shared environment unless they were very large. Weight differences between affected and unaffected cotwins from monozygotic (MZ) discordant pairs were large for T2D and GD, but much larger again for discordant dizygotic (DZ) pairs. The bivariate genetic analysis (under the multifactorial threshold model) estimated the genetic correlation between body mass index (BMI) and T2D to be 0.46, and the environmental correlation at only 0.06.