A selective estimation of glycosylated hemoglobin, HbA1C, exclusive of other minor components, is propably one of the best available tests. Such an automated method is proposed, which was experimented in 550 cases. It demonstrates a quite narrow gaussian distribution in controls. In contrast, a large dispersion is observed among long term diabetics, most of them with ocular complications with only a minority of results in the normal range. This is still more evident in case of juvenile diabetes.