To enhance the safety and efficiency of membranous nephropathy (MN) detection, surface-enhanced Raman spectroscopy (SERS) was studied for the detection of biological samples. SERS was used to detect urine samples from 60 patients with membranous nephropathy (MN), 84 patients with IgA nephropathy (IgAN), 26 patients with diabetic nephropathy (DN) and 68 patients with glomerular disease (GD). A SERS special library of MN, IgAN, DN and GD was established. The data of the SERS library were divided into three groups (MN/IgAN, MN/DN, MN/GD). The Raman characteristic peaks of spectral data were compared and analyzed to clarify their assignment and biomedical significance. On this basis, the classification model was established by principal component analysis?linear discriminant analysis (PCA?LDA) and partial least squares discriminant analysis (PLS?DA). The results show that the PCA?LDA?based classification model has a classification accuracy of 100.00% for MN/IgAN, MN/DN and MN/GD. The classification accuracy of the PLS?DA?based classification model for MN/IgAN, MN/DN and MN/GD is 95.12%, 100.00% and 96.09%, respectively. The PCA?LDA model has higher accuracy, and the accuracy of the optimized model for MN/IgAN, MN/DN and MN/GD classification is 74.38%, 90.70% and 77.34%, respectively. Therefore, SERS has potential for clinical diagnosis in differentiating MN patients from other patients with kidney disease.