The aim was to assess the long-term effect of a loose restriction of carbohydrate intake (carbohydrate-reduced diet: CARD) compared to a conventional diet (CD) in type 2 diabetes. One hundred and thirty-three type 2 diabetic outpatients followed the CD (n=57, 1734+/-410 kcal, carbohydrate:protein:fat ratio=57:16:26) or CARD (n=76, 1773+/-441 kcal, carbohydrate:protein:fat ratio=45:18:33) according to their own will, and were followed up for 2 years. Glycemic control, body mass index (BMI), serum cholesterols and dose of antidiabetic drugs were assessed at baseline and after 1 and 2 years. At baseline, hemoglobin A1c (HbA1c) and BMI levels were 7.1+/-1.0% and 24.2+/-2.9, respectively, in the CD group, and 7.4+/-1.1% and 25.1+/-3.4 in the CARD group, showing no significant differences. During the 2-year follow-up period, HbA1c levels were significantly improved in the CARD group (CD: 7.5+/-1.3%, CARD: 6.7+/-0.6%, P<0.001), and BMI decreased more significantly in the CARD group (CD: 23.8+/-3.0, CARD: 23.8+/-3.5, P<0.001). The doses of sulfonylureas clearly tapered, and serum cholesterol profiles improved significantly with the CARD. Our results warrant a long-term and large-scale randomized study of the diet for type 2 diabetes.