We conducted a meta-analysis of relevant cohort studies to investigate the relationships between cyclooxygenase-2 (COX-2) protein and the prognosis of pancreatic cancer. The following electronic databases were searched without language restrictions: MEDLINE (1966∼2013), the Library Database (Issue 12, 2013), EMBASE (1980∼2013), CINAHL (1982∼2013), Web of Science (1945∼2013), and the Chinese Biomedical Database (CBM) (1982∼2013). Meta-analysis was performed using the STATA statistical software. Six cohort studies with a total of 712 pancreatic cancer patients were involved in this meta-analysis. Our findings showed that COX-2-positive patients were significantly associated with a shorter overall survival (OS) than COX-2-negative patients (hazard ratio (HR) = 1.48, 95 % confidence interval (95%CI) = 1.12∼1.85, P < 0.001). A subgroup analysis by ethnicity also revealed that pancreatic cancer patients with an abnormal COX-2 expression exhibited a worse OS than COX-2-negative patients among both Asians and Caucasians (Asians: HR = 1.40, 95%CI = -0.09∼2.89, P = 0.066; Caucasians: HR = 1.49, 95%CI = 1.11∼1.87, P < 0.001, respectively). Our findings provide empirical evidence that abnormal COX-2 expression may be strongly correlated with poor prognosis for patients with pancreatic cancer. Thus, COX-2 protein may be a useful biomarker for pancreatic cancer.