The frequency distributions of both the fasting and two-hour post-load plasma glucose levels were bimodal in the Pima Indian population aged 25 years and over. The hyperglycaemic component of this distribution represents those with diabetes mellitus, as some 30 percent of this group had evidence of the specific vascular complications of the disease, whereas these abnormalities were virtually absent in those with lower glucose levels. The bimodal characteristics of the frequency distributions were utilized to define optimal criteria to separate those with and without diabetes. The sensitivity and specificity of these criteria for fasting and two-hour glucose levels were compared and were found to be similar. The fasting glucose determination, however, was more reproducible and stable, as well as being easier to obtain, indicating that it is the better measurement for diagnostic purposes. The optimal level for diagnosis of 7.5 mmol/l (136 mg/dl) for the fasting glucose and the equivalent two-hour value of 14 mmol/l (250 mg/dl), were higher than many previously recommended diagnostic levels. Nevertheless, there was no evidence that subjects with lower levels were at appreciable risk of developing the specific complications of diabetes. Subjects with impaired glucose tolerance (IGT), but without fasting hyperglycaemia, should not be diagnosed as having diabetes mellitus.