OBJECTIVE: To define the threshold for excess glucose variability (GV), one of the main features of dysglycemia in diabetes. RESEARCH DESIGN AND METHODS: A total of 376 persons with diabetes investigated at the University Hospital of Montpellier (Montpellier, France) underwent continuous glucose monitoring. Participants with type 2 diabetes were divided into several groups-groups 1, 2a, 2b, and 3 (n = 82, 28, 65, and 79, respectively)-according to treatment: 1) diet and/or insulin sensitizers alone; 2) oral therapy including an insulinotropic agent, dipeptidyl peptidase 4 inhibitors (group 2a) or sulfonylureas (group 2b); or 3) insulin. Group 4 included 122 persons with type 1 diabetes. Percentage coefficient of variation for glucose (%CV = [(SD of glucose)/(mean glucose)] x 100) and frequencies of hypoglycemia (interstitial glucose <56 mg/dL [3.1 mmol/L]) were computed. RESULTS: Percentages of CV (median [interquartile range]; %) increased significantly (P < 0.0001) from group 1 (18.1 [15.2-23.9]) to group 4 (37.2 [31.0-42.3]). In group 1, the upper limit of %CV, which served as reference for defining excess GV, was 36%. Percentages of patients with %CVs above this threshold in groups 2a, 2b, 3, and 4 were 0, 12.3, 19.0, and 55.7%, respectively. Hypoglycemia was more frequent in group 2b (P < 0.01) and groups 3 and 4 (P < 0.0001) when subjects with a %CV >36% were compared with those with %CV </=36%. CONCLUSIONS: A %CV of 36% appears to be a suitable threshold to distinguish between stable and unstable glycemia in diabetes because beyond this limit, the frequency of hypoglycemia is significantly increased, especially in insulin-treated subjects.