OBJECTIVE - In a previous study, we found observational evidence of improvement in β-cell function when rosiglitazone was added to a failing oral antihyperglycemic regimen consisting of sulfonylureas and metformin. To confirm our previous observations, we designed and performed a prospective, randomized, and controlled study. RESEARCH DESIGN AND METHODS - A total of 17 subjects with type 2 diabetes, inadequately controlled on a maximized oral antihyperglycemic double regimen of glimepiride and metformin, were randomized to the addition of rosiglitazone or insulin to their treatment regimens for a period of 6 months. At baseline and at 6 months, the following were performed: measurement of fasting plasma glucose, fasting proinsulin, and insulin levels; frequently sampled intravenous glucose tolerance test; and glucagon stimulation test for C-peptide. RESULTS - Nine subjects were randomized to the addition of 8 mg rosiglitazone, and eight subjects were randomized to the addition of one injection of insulin (premixed 70/30) before their evening meal. The treatment groups were well matched for age, duration of diabetes, and BMI. Most important, the HbA1c was well matched between groups before treatment (8.7 ± 0.3 and 9.0 ± 0.3%; NS) and at the end of the 6 months (7.8 ± 0.5 and 7.8 ± 0.3%; NS). After 6 months, at the end of the study, there was a significant improvement in acute insulin response to glucose in the rosiglitazone group (+15.3 μIU · ml-1 · 10 min-1; P < 0.001) that led to an increase in the disposition index from 0.18 at baseline to 4.18 at 6 months (P = 0.02). Furthermore, at the end of the study, the proinsulin-to-insulin ratio had decreased in the rosiglitazone group by 36% (P = 0.03) but did not change significantly in the insulin treatment group. CONCLUSIONS - Rosiglitazone, but not insulin, induced a recovery of pancreatic β-cell function, as evidenced by the restoration of the first-phase insulin response to glucose, improvement in the disposition index, and a decrease in the proinsulin-to-insulin ratio in subjects with type 2 diabetes in whom oral antihyperglycemic therapy failed. This improvement was independent of the correction of glucotoxicity.