Summary Background The evolving epidemic of type 2 diabetes has challenged health-care providers to assess the safety and efficacy of various diabetes prevention strategies. The CANOE (CAnadian Normoglycemia Outcomes Evaluation) trial investigated whether low-dose combination therapy would affect development of type 2 diabetes. Methods In this double-blind, randomised controlled trial undertaken in clinics in Canadian centres, 207 patients with impaired glucose tolerance were randomly assigned to receive combination rosiglitazone (2 mg) and metformin (500 mg) twice daily or matching placebo for a median of 3·9 years (IQR 3·0–4·6). Randomisation was computer-generated in blocks of four, with both participants and investigators masked to treatment allocation. The primary outcome was time to development of diabetes, measured by an oral glucose tolerance test or two fasting plasma glucose values of 7·0 mmol/L or greater. Analysis was by intention to treat. This study is registered with ClinicalTrials.gov, number NCT00116932. Findings 103 participants were assigned to rosiglitazone and metformin, and 104 to placebo; all were analysed. Vital status was obtained in 198 (96%) participants, and medication compliance (taking at least 80% of assigned medication) was 78% (n=77) in the metformin and rosiglitazone group and 81% (n=80) in the placebo group. Incident diabetes occurred in significantly fewer individuals in the active treatment group (n=14 [14%]) than in the placebo group (n=41 [39%]; p<0·0001). The relative risk reduction was 66% (95% CI 41–80) and the absolute risk reduction was 26% (14–37), yielding a number needed to treat of 4 (2·70–7·14). 70 (80%) patients in the treatment group regressed to normal glucose tolerance compared with 52 (53%) in the placebo group (p=0·0002). Insulin sensitivity decreased by study end in the placebo group (median −1·24, IQR −2·38 to −0·08) and remained unchanged with rosiglitazone and metformin treatment (−0·39, −1·30 to 0·84; p=0·0006 between groups). The change in β-cell function, as measured by the insulin secretion-sensitivity index-2, did not differ between groups (placebo −252·3, −382·2 to −58·0 vs rosiglitazone and metformin −221·8, −330·4 to −87·8; p=0·28). We recorded an increase in diarrhoea in participants in the active treatment group compared with the placebo group (16 [16%] vs 6 [6%]; p=0·0253). Interpretation Low-dose combination therapy with rosiglitazone and metformin was highly effective in prevention of type 2 diabetes in patients with impaired glucose tolerance, with little effect on the clinically relevant adverse events of these two drugs. Funding GlaxoSmithKline.