Previously-proposed rheumatoid arthritis (RA) HLA-DRB1 susceptibility and protective models were compared, based on amino acids at positions 67-74 and autoantibody combinations. 3 657 RA patients and 1 357 controls were studied using logistic regression, with secondary stratification by anti-citrullinated peptide antibodies(ACPA) and rheumatoid factor(RF). Susceptibility models were based on previously defined HLA-DRB1 shared epitope(SE) subgroups. (70)DERAA(74), D(70) and I(67) protective models were compared, adjusting for HLA-DRB1 SE. A hierarchy of risk was observed within the HLA-DRB1 SE, particularly for ACPA-positive and RF-positive RA: HLA-DRB1(*)0401∼(*)0404>(*)0101∼(*)1001 ((*)0404>(*)0101: P=0.0003). HLA-DRB1(*)0401/(*)0404 compound heterozygosity conferred a risk similar to (*)0401 homozygosity (P=0.70). Protective effects of D(70) and I(67) were similar. Predictions of the D(70) model fitted the data better than those of the I(67) model. The protective effect of D(70) showed a gene-dose effect (OR 0.82, 95% CI 0.73-0.92, P=5.8 × 10(-4)), but was only seen in RA patients positive for RF or ACPA. HLA-DRB1 SE alleles were also associated with ACPA-negative, RF-positive RA (OR 1.42 (1.15-1.76), P=0.0012). In conclusion, HLA-DRB1 SE alleles show heterogeneity in RA susceptibility; their major effect appears to be mediated by ACPA positivity, but a significant association of HLA-DRB1 SE with RF-positive, ACPA-negative RA was also observed. D(70) specifically protected against antibody-positive RA.