The effect of the start of glycemic control on the progression of retinopathy was investigated by a case-control study. The changes in glycosylated hemoglobin (HbA1) were compared between a case group, diabetic cases showing progression of retinopathy (Group 1), and a control group, diabetic cases showing no progression of retinopathy (Groups 2-A and 2-B). Group 2-A was matched with Group 1 on the basis of the grade of retinopathy at the first examination and other clinical data. Group 2-B was matched with Group 1 in terms of HbA1 value and methods of control, but had no retinopathy or background retinopathy. The retrospective follow-up period for the three groups was 24 months. On the basis of the respective matching factors, Groups 1 and 2-A were divided into 9 blocks of homogeneous subjects, and Groups 1 and 2-B were similarly divided into 6 blocks. The resulting data was evaluated block by block, using the analysis of variance (ANOVA) and a conditional logistic regression analysis. In Group 1, the HbA1 value decreased rapidly 10-6 months before the progression of retinopathy, but the HbA1 value did not change in Groups 2-A and 2-B during the 24-month follow-up. The difference in the estimated mean HbA1 value between 10-9 months and 1-0 month before the progression of retinopathy was 2.46% greater in Group 1 than in Group 2-A, as determined by ANOVA. The relative risks of a 1, 2 and 3% increase in HbA1 value for 7-6 months were estimated as 1.6, 2.4 and 3.8, respectively, by conditional logistic regression analysis. These findings indicate that the decrease in HbA1 value during any 6-month period should be limited to less than 2% in order to prevent the progression of retinopathy. It is also evident that too rapid a decrease at the initiation of glycemic control could cause severe or transient exacerbation of the progression of retinopathy.