Aims/hypothesis Supervised exercise programmes improve glycaemic control in type 2 diabetes, but training characteristics associated with reduction in HbA 1c remain unclear. We conducted a systematic review with meta-regression analysis of randomised clinical trials (RCTs) assessing the association between intensity and volume of exercise training (aerobic, resistance or combined) and HbA 1c changes in patients with type 2 diabetes. Methods Five electronic databases were searched to retrieve RCTs of at least 12 weeks' duration, consisting of supervised exercise training vs no intervention, that reported HbA 1c changes and exercise characteristics. Two independent reviewers conducted study selection and data extraction.Results Twenty-six RCTs (2,253 patients) met the inclusion criteria. In multivariate analysis, baseline HbA 1c and exercise frequency explained nearly 58% of between-study variance. Baseline HbA 1c was inversely correlated with HbA 1c reductions after the three types of exercise training. In aerobic training, exercise volume (represented by frequency of sessions) was associated with changes in HbA 1c (weighted r0−0.64), while no variables were correlated with glycaemic control induced by resistance training. In combined training, weekly volume of resistance exercise explained heterogeneity in multivariate analysis and was associated with changes in HbA 1c levels (weighted r0−0.70). Conclusions/interpretation Reduction in HbA 1c is associated with exercise frequency in supervised aerobic training, and with weekly volume of resistance exercise in supervised combined training. Therefore, exercise volume is a major determinant of glycaemic control in patients with type 2 diabetes.