Background Early recognition and treatment of autoimmune encephalitis (AE) are crucial for patients, but diagnosis is often difficult and time-consuming. For this purpose, a syndrome-based diagnostic approach was published by Graus et al. (Lancet Neurol 15:391–404, 2016), but very little is known in the literature about its application in clinical practice. Aim Our aims are to test the feasibility of such approach in a real-world single-centre setting and to analyse the most relevant factors in criteria fulfilment. Methods We retrospectively applied these criteria to our cohort of patients discharged from our hospital with diagnosis of autoimmune encephalitis ( n  = 33, 58% antibody-positive). Results All the subjects fulfilled criteria for possible AE (pAE), with EEG and MRI playing a central role in diagnosis, while CSF was useful mainly to rule out other conditions. Three patients respected criteria for probable anti-NMDA-R encephalitis (pNMDA). Definite anti-NMDAR encephalitis was diagnosed in 4 patients with detection of the autoantibody but, surprisingly, none of these subjects had fulfilled criteria for pNMDA. 18 patients were diagnosed with definite limbic AE (15 patients were antibody-positive, three antibody-negative). Need for MRI bilateral involvement in antibody-negative limbic AE limited diagnosis. One patient fulfilled criteria for probable antibody-negative AE, while ten patients remained classified as pAE. Conclusion From our retrospective analysis, some suggestions for a better definition of the criteria may emerge. Larger studies on prospective cohorts may be more helpful to explore possible important issues.