To assess the extent to which the progression of diabetic retinopathy can be arrested by improved glycemic control, 35 normal dogs were randomly divided into a nondiabetic and three alloxan-induced diabetic groups prospectively identified according to glycemic control: poor control for 5 yr (PC), good control for 5 yr (GC), and poor control for 2.5 yr followed by good control for 2.5 yr (PGC). To achieve good control, insulin was given twice daily together with a measured diet so that hyperglycemia and glucosuria were mild and infrequent, and HbA1 was comparable to normal. Retinal capillary aneurysms and other lesions developed during 60 mo of poor control (group PC) and were inhibited if good control was begun promptly within 2 mo (group GC). In group PGC, retinopathy was absent or equivocal at 2.5 yr of poor control and, surprisingly, was found to develop subsequently despite good glycemic control. Retinopathy in group PGC was greater at autopsy than at 2.5 yr and was greater than in group GC. The results indicate that retinopathy may be preventable but tends to resist arrest even in its incipient stages, before more than the first few aneurysms have appeared.