We assessed the interobserver agreement on the diagnosis of multiple sclerosis (MS) in a study sample consisting of 41 MS (15 relapsing remitting, two secondary progressive, five primary progressive and 19 presenting their first clinical attack) and three non-MS cases. Clinical and paraclinical information was recorded in standardized forms. Four neurologists were asked to make a diagnosis using Poser's and McDonald's criteria and to assess MRI scans according to the McDonald's guidelines. In terms of the kappa statistic (kappa), we found a moderate agreement on the overall diagnosis using both Poser's and McDonald's criteria (kappa, respectively 0.57 and 0.52). As for distinct diagnostic categories, we observed a moderate to substantial agreement for the three McDonald categories (range of kappa values 0.49-0.64) and a fair to substantial agreement for the nine Poser categories (range of kappa values 0.37-0.67). Taking into account clinical information, the agreement on dissemination over time was substantially higher (kappa = 0.69) than that found on dissemination over space (kappa = 0.46). In contrast, for MRI assessment, the agreement for spatial dissemination was substantial (kappa = 0.74) compared with the fair agreement (kappa = 0.25) yielded by dissemination over time. The new McDonald's criteria yield a good overall diagnostic reliability, and compare favourably with Poser's classification in terms of agreement on distinct diagnostic categories.