OBJECTIVE: To explore the accuracy of plasma neurofilament light chain (NfL) as a biomarker for diagnosis and staging of cognitive impairment, in a large cohort with of previously diagnosed patients in clinical practice. METHODS: Retrospective, cross-sectional, monocentric study, from a tertiary memory clinic. Patients underwent cerebrospinal fluid core Alzheimer's disease (AD) biomarker evaluation using ELISA or Elecsys methods, and plasma NfL analysis using the single molecule array technology. The patients' biomarker data were examined for associations with: i/cognitive status ii/presence of neurodegenerative disease and iii/diagnostic groups. Associations between core CSF biomarkers and plasma NfL were determined. RESULTS: Participants (N = 558, mean age = 69.2 ± 8.8, 56.5% women) were diagnosed with AD (n = 274, considering dementia and MCI stages), frontotemporal dementia (FTD, n = 55), Lewy body disease (LBD, n = 40, considering MCI and dementia stages), other neurodegenerative diseases, n = 57 (e.g Supranuclear Palsy, Corticobasal syndrome), non-neurodegenerative cognitive impairment (NND, n = 79, e.g. vascular lesions, epilepsy or psychiatric disorders) or subjective cognitive impairment (SCI, n = 53). Mean plasma NfL (log, pg/mL) levels were higher in neurodegenerative than non-neurodegenerative disorders (1.35 ± 0.2 vs 1.16 ± 0.23, p < 0.001), higher in all diagnostic groups than in SCI (1.06 ± 0.23) p < 0.001), and associated with the stage of cognitive impairment (p < 0.001). The addition of plasma NfL to a clinical model (age, MMSE and APOE ε4 carriership) marginally improved the discrimination of degenerative from non-degenerative disorders in ROC analysis (AUC clinical model: 0.81, 95% CI = [0.77;0.85] AUC clinical model + plasma NfL: AUC = 0.83 95% CI = [0.78;0.87], delta Akaike information criterion = -11.7). DISCUSSION: Plasma NfL could help discrimination between degenerative and non-degenerative cognitive disorders, albeit not better than comprehensive clinical evaluation.