Background: Our study was designed to improve the accuracy of determining whether pulmonary nodules are benign or malignant.Methods: We evaluated the clinical and imaging features and serum markers: neuron specific enolase (NSE), carcino-embryonic antigen (CEA), cytokeratin fragment antigen 21-1 (CYFRA 21-1), miRNA-21-5p, and miR-574-5pof in 39 patients with pathology information. Factors that differed significantly between those with benign versus malignant pulmonary nodules were used to establish a prediction model for identifying malignant nodules.Results: The studied nodules were 51.3% malignant and 48.7% benign. Age, smoking status, nodule diameter, history of emphysema, vascular sign, burr sign, CYFRA21-1, CEA, miRNA-21-5p, and miRNA-574-5p differed significantly between the benign and malignant nodule groups. Serum levels of CYRFA21-1 and CEA could be used to distinguish between malignant and benign nodules with a positive predictive value (PPV) of 80.0%, a negative predictive value (NPV) of 84.2%, and an area under the receiver operating characteristics curve (AUC) of 0.863. Using the serum levels of miRNA-21-5p and miRNA-574-5p, the PPV was 55%, the NPV was 84.2%, and the AUC was 0.797. When all four serum markers were combined, the PPV was 80%, the NPV was 89.5%, and the AUC was 0.921. We established a prediction model for malignant nodules, including clinical features, imaging features, and serum markers. In cross-validation, the ratio of discriminant conformance was 95%.Conclusions: Serum levels of miRNA-21-5p and miRNA-574-5p are significantly higher in patients with malignant nodules than in patients with benign nodules and are potential serum biomarkers. Our prediction model could improve malignant nodule diagnosis.