Type 2 diabetes (T2D) is a highly prevalent but preventable disorder. We assessed the association between an a priori Mediterranean diet (MeDiet) score and fasting glucose and insulin at baseline and incident T2D after a 6-year follow-up in the Multi-Ethnic Study of Atherosclerosis. Dietary intake was measured at baseline using a 127-item FFQ in 5390 men and women aged 45–84 years free of prevalent diabetes and clinical CVD. A MeDiet score was created based on the intake of ten food components: vegetables; whole grains; nuts; legumes; fruits; ratio of monounsaturated:saturated fat; red and processed meat; dairy products; fish; alcohol. Multivariable linear and proportional hazards models were used to estimate the association of the MeDiet, categorised in quintiles, with baseline insulin and glucose, and incident diabetes, respectively. The models were adjusted for demographic, physiological and behavioural characteristics. After multivariable adjustment, individuals with a higher MeDiet score had lower baseline mean insulin levels (Q1: 5·8 (95 % CI 5·6, 6·0) μmol/l; Q5: 4·8 (95 % CI 4·6, 5·0) μmol/l; P for trend < 0·0001). A higher MeDiet score was also associated with significantly lower glucose levels after basic adjustment, but was attenuated after adjustment for waist circumference. During the follow-up, 412 incident diabetes events accrued. The MeDiet was not significantly related to the risk of incident diabetes (P for trend = 0·64). In summary, greater consistency with a Mediterranean-style diet, reflected by a higher a priori MeDiet score, was cross-sectionally associated with lower insulin levels among non-diabetics, and with lower blood glucose before adjustment for obesity, but not with a lower incidence of diabetes.