Increasing studies demonstrate the significant contributions of circRNA-related competitive endogenous RNA (ceRNA) regulatory networks to tumorigenesis and cancer progression. Here, we aimed to construct a non-small cell lung cancer (NSCLC)-specific circRNA-miRNA network and evaluate its diagnostic potential in NSCLC. MiRNA deep sequencing was performed to screen differentially-expressed serum miRNAs in NSCLC. Four bioinformatics databases (TargetScan, miRanda, starBase, and RNAhybrid) were used to analyze the integrated circRNA-miRNA interaction network. The circRNA-miRNA network, including hsa-miR-4482-3p, hsa-miR-146a-3p, hsa_circ_0008167 and hsa_circ_0003317 was constructed based on their interactions and preliminary testing in NSCLC cells. The relative levels of the selected non-coding RNAs (ncRNAs) were quantified using quantitative real-time polymerase chain reaction (qRT-PCR) in the healthy, pneumonia, benign lung tumor and NSCLC cohorts. The diagnostic power of the circRNA-miRNA network was evaluated using receiver operating characteristic (ROC) analyses. The serum levels of hsa-miR-4482-3p, hsa-miR-146a-3p, hsa_circ_0008167, and hsa_circ_0003317 were dysregulated in NSCLC. The combination of the four ncRNAs showed the highest diagnostic value to discriminate between benign lung tumors and NSCLC. Additionally, the upregulated levels of hsa_circ_0008167 were correlated to more aggressive features of NSCLC, such as lymph node metastasis, distant metastasis, and higher stage. Furthermore, the combination of hsa_circ_0008167 + hsa-miR-4482-3p, and hsa_circ_0008167 + hsa-miR-4482-3p + hsa-miR-146a-3p had the greatest diagnostic power to differentiate between lymph node +/- metastases and higher/lower stages, respectively, compared to circRNAs or miRNAs alone, and traditional tumor markers. In conclusion, we identified a specific circRNA-miRNA network with higher sensitivity and specificity to diagnose NSCLC, thereby providing a new strategy for further development of ceRNA-related tumor markers in other cancers. KEY MESSAGES: Serum miR-4482-3p, miR-146a-3p, circ_0008167 and circ_0003317 are dysregulated in NSCLC. Higher levels of serum circ_0008167 are associated with more malignant NSCLC. Multiple combinations of circRNAs and miRNAs show higher value to diagnose NSCLC.