OBJECTIVE To determine whether HLA exerts a variable influence on the predisposition of siblings of probands with clinically mild and severe rheumatoid arthritis (RA). METHODS Calculation of crude and adjusted odds ratios for concordance rates in sibships sharing two, one and no HLA haplotypes with a proband with clinically mild and severe RA, and HLA haplotype sharing in multiply affected sibships in the same clinical groups. RESULTS Compared with a reference value of 1.0 in siblings sharing no HLA haplotypes with a proband with mild RA, siblings sharing two HLA haplotypes with a severely affected proband had a sibship concordance rate odds ratio of 9.7 (95% confidence interval 2.5 to 38.2). When adjusted for age, sex, and disease duration, the odds ratio was 7.6 (1.8 to 32.4). No other sibships showed concordance rates which were significantly higher than the reference group. HLA haplotype sharing in multiply affected sibships in which the proband had severe RA deviated significantly from random (two, one, and no HLA haplotypes shared: 53.3, 40, and 6.7%, respectively; expected 25, 50, and 25%), whereas in sibships of probands with mild RA they did not (14.6, 70.8, and 14.6%). CONCLUSIONS In the predisposition of siblings to RA, sharing HLA haplotypes with a proband is only important if the proband has severe RA. Mild RA is not genetically linked to the HLA region.