Objectives: Carcinoembryonic antigen (CEA) is the most frequently used tumor marker for non-small cell lung cancer (NSCLC). The current study aimed to provide the highest-level evidence of the prognostic value of pretreatment serum CEA level for NSCLC through the appropriate statistical methodology and large-sample cohorts. Methods: The current retrospective cohort study with 1130 patients with NSCLC treated by thoracic surgery with pretreatment serum CEA concentrations above/below 5 ng/mL. Propensity score matching, Kaplan-Miere survival analysis, and Cox proportional hazard regression models were used to study the intergroup variance. The overall/disease-free hazard ratios (HRs) of the current study were combined with the previously published studies using cumulative meta-analysis to provide the highest-level evidence. Results: Intergroup confounding variables were well controlled by propensity score matching, and the survival differences were statistically significant. The Cox univariate analysis showed that the overall and disease-free HRs of the high CEA towards patients with low CEA were 1.595 (95% CI: 1.329-1.863, P = 0.004) and 1.498 (95% CI: 1.271-1.881, P = 0.004). The HRs of multivariate analysis were adjusted to 1.586 (95% CI: 1.398-1.812, P = 0.016) and 1.413 (95% CI: 1.22-1.734, P = 0.022) respectively. The cumulative meta-analysis showed that the cumulative overall HR was in accord with previous studies, and the cumulative disease-free HR turn to be statistically significant. Conclusions: Pretreatment serum CEA level was an independent influence factor of overall/disease-free survival of patients with NSCLC, and even for patients with the same pTNM stages or pathologic stages, it is used for prognosis.