The patients of a multiple sclerosis (MS) incidence cohort with 25 years of longitudinal follow-up were typed for HLA-DR and DQ. This type of cohort provides reliable data for gene frequencies and prognostic studies. The influence of sampling bias, mainly due to mortality during the long follow-up, was accounted for. A positive association between MS and DR15,DQ6 was confirmed, but this haplotype did not influence prognosis. There was no difference in haplotype frequency between relapsing-remitting and primary chronic progressive MS. DR17,DQ2 was significantly over-represented in the quartile with the most malignant course. The haplotype DR1,DQ5, which was found rather less frequently in MS patients, also tended to be associated with a poorer prognosis.