We conducted a meta-analysis to analyse the effect of metformin on survival of pancreatic cancer patients at various stages. We performed a systematic search of PubMed, Embase, Cochrane, and Web of Science to identify all relevant studies. Summary hazard ratios (HR) of survival and 95% confidence intervals (95% CI) were calculated with a fixed or random effects model according to inter-study heterogeneity. Nine retrospective cohort studies and two randomized controlled trials (RCTs) were eligible. There was a significant improvement in survival (HR = 0.86, 95% CI 0.76-0.97; P < 0.05) in the metformin group compared with control. Subgroup analysis indicated that metformin improved survival in patients with resection (HR = 0.79, 95% CI 0.69-0.91; P < 0.05) and patients with locally advanced tumors (HR = 0.68, 95% CI 0.55-0.84; P < 0.05) but not in patients with metastatic tumors, even when RCT data were included (HR = 0.99, 95% CI 0.70-1.40; P > 0.05), or were excluded (HR = 0.89, 95% CI 0.61-1.31; P > 0.05). This meta-analysis indicated that the effect of metformin does correlate with tumor stage but should be prudently considered given the limited and variable studies performed to data.