Interobserver agreement in the clinical diagnosis of multiple sclerosis (MS) among six neurologists was evaluated. Three of them participated in a study of the clinical diagnosis of MS, the Italian Multicenter Study (IMS). The raters examined the clinical forms of MS of 50 patients randomly selected from among 430 patients recruited from the IMS. For each patient, neurologists were asked to make a diagnosis according to the McDonald-Halliday classification system of MS. The overall agreement on the diagnosis (MS present or absent) was fair, with no difference noted between the two groups of raters. Considering the six diagnostic levels instead, the reliability was higher for the neurologists participating in the IMS program. These neurologists agreed particularly on the Clinically Definite and Progressive Possible classifications. Complete disagreement was observed for the Early Probable or Latent and Progressive Probable classifications. Because of the different level of agreement on diagnosis, we suggest separate consideration of Clinically Definite and Progressive Probable MS cases in clinical trials and epidemiologic studies of this disease.