To determine whether high-ketogenic very-low-energy diets (VLEDs) can reduce hepatic glucose output (HGO) and hyperglycemia more effectively than can low-ketogenic VLEDs in obese patients with non-insulin-dependent diabetes mellitus (NIDDM), seven patients were treated with a high-ketogenic VLED for 3 wk and were compared with six patients treated with a low-ketogenic VLED. All patients were then crossed over and treated with the alternate diet for another 3 wk. Basal HGO, fasting ketone bodies, and glycemia, insulin, and C-peptide after fasting and an oral-glucose-tolerance test (OGTT) were measured. Before treatment, prediet weight and fasting, OGTT, and HGO measurements were not different between groups. After dieting, weight loss was not different between the groups. However, fasting and OGTT glycemia were lower during treatment with the high-ketogenic VLED than with the low-ketogenic VLED (treatment effect: P < 0.05, by analysis of variance). Moreover, there was a strong correlation between basal HGO and fasting plasma ketone bodies (r = -0.71 at 3 wk, r = -0.67 at 6 wk; both P < 0.05). In contrast, fasting and OGTT plasma insulin and C-peptide concentrations were not different between treatment groups. These data indicate that in obese patients with NIDDM, high-ketogenic VLEDs have a more clinically favorable effect on glycemia than do low-ketogenic VLEDs.