Objective To investigate the value of surface-enhanced Raman spectroscopy (SERS) of peripheral erythrocytes in the diagnosis of geriatric cognitive disorder. Methods A total of 62 patients (aged >65 years) in the Department of Geriatrics of the Sixth People′s Hospital recieved neuropsychological and clinical assessment. Accordingly, they were divided into normal control (NC) group, mild cognitive impairment (MCI) group, and Alzheimer disease (AD) group. The baseline characteristics were collected and recorded. SERS was performed for erythrocytes. A fifth-order polynomial fitting algorithm was employed to process and extract effective SERS data, linear discriminant method to build a data model, and receiver operating characteristic curve to examine diagnostic value. Results Differences between groups were observed in the SERS of erythrocytes. Discriminant model for NC and AD groups can accuratly distinguish 71.8% patients with a sensitivity of 91.3% and specificity of 56.2%, discriminant model for NC and MCI groups can distinguish 58.7% patients with a sensitivity of 56.5% and specificity of 73.9%, and discriminant model for AD and MCI groups can accuratly distinguish 61.5% patients with a sensitivity of 75.0% and specificity of 65.2%. Conclusion SERS of erythrocytes proves to be of diagnostic value for cognitive impairment, inspiring the research on pathophysiology of erythrocytes in AD.