OBJECTIVE--To determine the prevalence of diabetes in relation to birth weight in Pima Indians. DESIGN--Follow up study of infants born during 1940-72 who had undergone a glucose tolerance test at ages 20-39 years. SETTING--Gila River Indian community, Arizona. SUBJECTS--1179 American Indians. MAIN OUTCOME MEASURE--Prevalence of non-insulin dependent diabetes mellitus (plasma glucose concentration > or = 11.1 mmol/l two hours after ingestion of carbohydrate). RESULTS--The prevalence was greatest in those with the lowest and highest birth weights. The age adjusted prevalences for birth weights < 2500 g, 2500-4499 g, and > or = 4500 g were 30%, 17%, and 32%, respectively. When age, sex, body mass index, maternal diabetes during pregnancy, and birth year were controlled for, subjects with birth weights < 2500 g had a higher rate than those with weights 2500-4499 g (odds ratio 3.81; 95% confidence interval 1.70 to 8.52). The risk for subsequent diabetes among higher birthweight infants (> or = 4500 g) was associated with maternal diabetes during pregnancy. Most diabetes, however, occurred in subjects with intermediate birth weights (2500-4500 g). CONCLUSIONS--The relation of the prevalence of diabetes to birth weight in the Pima Indians is U shaped and is related to parental diabetes. Low birth weight is associated with non-insulin dependent diabetes. Given the high mortality of low birthweight infants selective survival in infancy of those genetically predisposed to insulin resistance and diabetes provides an explanation for the observed relation between low birth weight and diabetes and the high prevalence of diabetes in many populations.