Epidemiologic studies have observed association between short sleep duration and both cardiovascular disease (CVD) and type 2 diabetes, although these results may reflect confounding by pre-existing illness. This study aimed to determine whether short sleep duration predicts future CVD or type 2 diabetes after accounting for baseline health. Baseline data for 241,949 adults were collected through the 45 and Up Study, an Australian prospective cohort study, with health outcomes identified via electronic database linkage. Cox proportional hazards models were used to estimate hazard ratios (HR) and 95% confidence intervals. Compared to 7h sleep, <6h sleep was associated with incident CVD in participants reporting ill-health at baseline (HR=1·38 [95% CI: 1·12-1·70]), but not after excluding those with baseline illness and adjusting for baseline health status (1·03 [0·88-1·21]). In contrast, the risk of incident type 2 diabetes was significantly increased in those with <6h versus 7h sleep, even after excluding those with baseline illness and adjusting for baseline health (HR=1·29 [1·08-1·53], P=0.004). This suggests the association is valid and does not simply reflect confounding or reverse causation. Meta-analysis of ten prospective studies including 447,124 participants also confirmed an association between short sleep and incident diabetes (1·33 [1·20-1·48]). Obtaining less than 6 hours of sleep each night (compared to 7 hours) may increase type 2 diabetes risk by approximately 30%.