Background and purpose Three different sets of criteria have been proposed for the diagnosis of neuromyelitis optica spectrum disorder (NMOSD). The objective was to compare the specificity, sensitivity and diagnostic accuracy of the three different sets of NMOSD criteria in patients presenting with inflammatory disorders of the central nervous system suggestive of NMOSD. Methods For 236 suspected NMOSD patients referred for serum aquaporin-4 immunoglobulin G antibody (AQP4-IgG) testing between 2012 and 2014, the three sets of NMOSD criteria [1999, 2006 NMO criteria and 2015 International Panel for NMO Diagnosis (IPND) criteria] were applied and compared to the final diagnosis. Results Seventy-six patients fulfilled at least one set of criteria and 28 patients fulfilled all NMOSD sets of criteria. The final diagnosis was NMOSD in 66 cases, multiple sclerosis according to the MacDonald 2010 criteria in 85 cases and another diagnosis in 85 cases. The 2006 NMO criteria have the highest specificity (99%) and the 2015 IPND NMOSD criteria the highest sensitivity (97%). For the 1999, 2006 and 2015 IPND NMOSD criteria, the accuracy was respectively 82%, 87% and 97%. Conclusions Our study highlights the limitations of the first set of criteria that include the optico-spinal form of multiple sclerosis. The accuracy of NMOSD diagnostic criteria improved from 1999 to 2015. It confirms the increased performance of the last set of criteria which covers a larger spectrum of clinical presentation. This study raises some concerns for classifying patients with seronegative transverse myelitis or optic neuritis, and myelin oligodendrocyte glycoprotein (MOG) antibody- associated disease.