1. In a cross-sectional study, 32 insulin-treated diabetic patients had elevated low shear (27.1 vs 22.1 mPa s, P less than 0.05) and high shear blood compared with 10 non-diabetic controls. After correction to 45% packed cell volume, the abnormality had a tendency to be greater in patients with proliferative (mean low shear viscosity, 30.8 mPa s) than background (29.2 mPa s) or nil/minimal retinopathy (27.6 mPa s, 0.05 less than P less than 0.97, permutational trend test). 2. The fibrinogen levels were higher in the diabetic group (P less than 0.05) and correlated with the low shear blood viscosity (rs = 0.38, P less than 0.05). 3. In a prospective study, 74 insulin-treated diabetic patients with background retinopathy were randomized into two groups. Thirty-six patients were on attempted improved therapy (A group); in these the mean glycosylated haemoglobin (Hb A1c) fell within 1 year (11.6 to 10.1%, P less than 0.001). both the corrected low shear blood and plasma viscosity fell similarly (P less than 0.001). The fall was greater than in those patients who were kept on usual therapy (U group) and whose glycosylated haemoglobin did not change significantly (11.7 to 11.4% over the year. 4. The effect of diabetes on blood viscosity may not be a direct pathological factor, as the same increased viscosity would be produced by a mean increase of 1.7% in packed cell volume, compared with a population range of 14% packed cell volume.