To date only one published large series of human leukocyte antigen matching and liver allograft survival exists, and considerable confusion has arisen about the advantage or disadvantage of human leukocyte antigen matching. In the present study we have reinvestigated the relationship between human leukocyte antigen mismatch and graft survival in 466 first liver allografts, seeking to clarify the relationship between human leukocyte antigen and both acute rejection and the vanishing bile duct syndrome. In view of current criticism regarding the accuracy of serological tissue typing for human leukocyte antigen-DR, we have used both classic serology and restriction fragment length polymorphism analysis to ensure the accurate assignment of recipient DR types. In addition, we have used polymerase chain reaction amplification and allele-specific and sequence-specific oligonucleotide probes to retest the hypothesis that human leukocyte antigen class I1 matching may increase susceptibility to the vanishing bile duct syndrome. One-year graft survival was significantlylower in patients with zero or two human leukocyte antigen-A mismatches (52% and 63%, respectively) than in those with one human leukocyte antigen-A mismatch (69%) (p = 0.016 and p = 0.018). A similar effect of B mismatching was observed, with a 1-yr graft survival of 73% for those with one compared with 60% for those with two human leukocyte antigen-B mismatches. In contrast no correlation was found between DR mismatch and graft survival. Human leukocyte antigen class I matching appears to influence graft survival largely through the occurrence of acute rejection and the development of the vanishing bile duct syndrome. In those patients who underwent liver biopsies, 49% (28 of 57) without acute rejection had a single B mismatch compared with 32% (88 of 263) with acute rejection (2 = 5.5, p < 0.025). In addition, a complete human leukocyte antigen-A mismatch was more common in those who developed the vanishing bile duct syndrome than in those with normal graft function (54% vs. 38%;