HLA compatibility is a key factor for survival after unrelated hematopoietic stem cell transplantation (HSCT). HLA-A, -B, -C, -DRB1, and -DQB1 are usually matched between donor and recipient. By contrast, HLA-DPB1 mismatches are frequent, although it is feasible to optimize donor selection and DPB1 matching with prospective typing. Because classical DPB1 allele mismatches are often unavoidable, however, several biological models have been developed to predict the optimal DPB1 mismatch combination for less graft-versus-host disease (GVHD) and better overall survival. In 909 recipient/donor pairs, we analyzed the role of 3 biological models: T-cell epitopes (TCEs) based on the immunogenicity of DPB1, cell surface expression of DPB1 molecules based on a single-nucleotide polymorphism located in the 3′ untranslated region, and the Predicted Indirectly ReCognizable HLA Epitopes (PIRCHE) model based on the presentation of allogeneic peptides derived from mismatched HLA, compared with the classical allele mismatch. Matching for both DPB1 alleles remains the best option to prevent acute GVHD. In the situation of one DPB1 allele mismatch, the donor associated with the lowest acute GVHD risks is mismatched for an allele with a low expression profile in the recipient, followed by a permissive TCE3/4 mismatch and/or the absence of PIRCHE II potential against the recipient. In the context of 2 DPB1 mismatches, the same considerations apply for a permissive TCE3/4 mismatch and no PIRCHE II. By combining the biological models, the most favorable DPB1 constellation can be defined. This approach will help optimize donor selection and improve post-HSCT complications and patient prognosis.