We conducted a prospective multicenter randomized trial to determine both the feasibility of maintaining blood glucose control at differing levels and the effect of improved control on diabetic microangiopathy and albuminuria. Seventy patients with diabetes (low C-peptide level) with nonproliferative retinopathy were randomly assigned to continuous subcutaneous insulin infusion or unchanged conventional injection treatment. At entry, both groups had similar demographic, clinical, and glycemic characteristics. Over the succeeding eight months, mean 24-hour glucose concentrations (175 +/- 9 mg per deciliter) and glycosylated hemoglobin levels (10.0 +/- 0.3 per cent) remained elevated during conventional treatment but fell to nearly normal levels (117 +/- 6 mg per deciliter and 8.1 +/- 0.2 per cent, respectively) with continuous insulin infusion. The frequency of biochemical hypoglycemia (less than 40 mg of blood glucose per deciliter) was similar in both groups, but ketoacidosis occurred only during continuous infusion. The level of retinopathy, assessed from photographs, progressed in both groups. Continuous infusion was associated with slightly more deterioration, mainly because of the appearance of soft exudates and intraretinal microvascular abnormalities. In contrast, elevated albumin-excretion rates fell during continuous infusion but not during conventional treatment. We conclude that maintenance of differing levels of blood glucose is feasible in a multicenter trial and that a nearly normal blood glucose level for eight months does not retard progression of, and may initially worsen, established retinopathy. These preliminary observations indicate the need for longer trials (particularly of primary prevention).