Few studies have determined whether carotid artery intima-media thickness (IMT) is associated prospectively with risk of first ischemic stroke. In the Atherosclerosis Risk in Communities Study, carotid IMT, an index of generalized atherosclerosis, was defined as the mean of IMT measured by B-mode ultrasonography at six sites of the carotid arteries. The authors assessed the relation of mean IMT to stroke incidence over 6-9 years' follow-up (1987-1995) among 7,865 women and 6,349 men aged 45-64 years without prior stroke at baseline in four US communities. There were 90 incident ischemic stroke events for women and 109 for men. In sex-specific Cox proportional hazards models adjusting only for age, race, and community, the hazard rate ratios comparing extreme mean IMT values (> or =1 mm) to values less than 0.6 mm were 8.5 for women (95% confidence interval: 3.5, 20.7) and 3.6 for men (95% confidence interval: 1.5, 9.2). The relation was graded, and models with cubic splines indicated significant nonlinearity, with hazards increasing more rapidly at lower IMTs than at higher IMTs. Thus, models using linear IMT values substantially underestimate the strength of the association at lower IMTs. The strength of the association was reduced by the inclusion of putative stroke risk factors, but it remained elevated at higher IMTs. Hence, mean carotid IMT is a noninvasive predictor of future ischemic stroke incidence.