OBJECTIVE: To examine the relationship of retinal vascular caliber to incident diabetes in a population-based cohort. RESEARCH DESIGN AND METHODS: The Australian Diabetes, Obesity and Lifestyle (AusDiab) Study recruited adults aged 25+ years across Australia in 1999-2000, with a follow-up 5 years later in 2004-2005. Participants' glycemic status was classified using fasting plasma glucose (FPG) and 2-h oral glucose tolerance (2-h plasma glucose [2hPG]) tests. Diabetes was diagnosed if FPG was >or=7.0 mmol/l or 2hPG was >or=11.1 mmol/l. Retinal vascular caliber was measured from baseline retinal photographs using a computer-assisted program. RESULTS: Of the 803 participants without diabetes at baseline, 108 (13.4%) developed diabetes at follow-up: 7 (2.8%) of 246 participants with normal glucose tolerance, 9 (13.6%) of 66 participants with impaired fasting glucose, and 92 (18.7%) of 491 participants with impaired glucose tolerance. After multivariate analysis, participants with narrower retinal arteriolar caliber had a higher risk of diabetes (odds ratio 2.21 [95% CI 1.02-4.80], comparing smallest versus highest arteriolar caliber tertiles, P = 0.04 for trend). There was no association between retinal venular caliber and incident diabetes. CONCLUSIONS: Narrower retinal arteriolar caliber predicted risk of diabetes. These data provide further evidence that microvascular changes may contribute to the pathogenesis of diabetes.