We aimed at investigating the preventive role of exercise intervention during pregnancy, in high-risk women for gestational diabetes mellitus (GDM). We searched PubMed, CENTRAL, and Scopus for randomized controlled trials (RCTs) that evaluated exercise interventions during pregnancy on women at high risk for GDM. Data were combined with random effects models. Between study heterogeneity (Cochran's Q statistic) and the extent of study effects variability I.sup.2 with 95% confidence interval (CI) were estimated. Sensitivity analyses examined the effect of population, intervention, and study characteristics. We also evaluated the potential for publication bias. Among the 1,508 high-risk women who were analyzed in 9 RCTs, 374 (24.8%) 160 (21.4%) in intervention, and 214 (28.1%) in control group developed GDM. Women who received exercise intervention during pregnancy were less likely to develop GDM compared to those who followed the standard prenatal care (OR 0.70, 95%CI 0.52, 0.93; P-value 0.02) Q 10.08, P-value 0.26; I.sup.2 21% (95%CI 0, 62%. Studies with low attrition bias also showed a similar result (OR 0.70, 95%CI 0.51, 0.97; P-value 0.03). A protective effect was also supported when analysis was limited to studies including women with low education level (OR 0.55; 95%CI 0.40, 0.74; P-value 0.0001); studies with exercise intervention duration more than 20 weeks (OR 0.54; 95%CI 0.40, 0.74; P-value 0.0007); and studies with a motivation component in the intervention (OR 0.69, 95%CI 0.50, 0.96; P-value 0.03). We could not exclude large variability in study effects because the upper limit of I.sup.2 confidence interval was higher than 50% for all analyses. There was no conclusive evidence for small study effects (P-value 0.31). Our study might support a protective effect of exercise intervention during pregnancy for high-risk women to prevent GDM. The protective result should be corroborated by large, high quality RCTs.