A human skin explant model has been used to predict the clinical outcome and to study the immunopathology of human graft-versus-host disease (GVHD). Whether the model gives the same predictive effect for GVHD in different hematopoietic stem cell transplantation (HSCT) settings has not been assessed. It is also unknown whether the skin explant result reflects the known biological risk factors for clinical GVHD. In this study, the skin explant model was used to detect graft-versus-host reactions (GVHR) in vitro for 225 eligible patient/donor pairs. The predicted skin GVHR grade was correlated with the outcome of clinical GVHD, as well as HLA matching status, sex mismatches, and patient age. In sibling HSCT under either myeloablative or reduced-intensity conditioning, a significant correlation was observed between the predicted skin GVHR and clinical GVHD (P < .001 and P = .033, respectively). In HSCT using unrelated donors, the involvement of T-cell depletion led to a sharp increase in false-positive GVHR results, and no correlation was observed between the predicted skin GVHR and clinical GVHD. The skin GVHR grade correlated significantly with the HLA matching status (HLA-matched sibling pairs, HLA-matched unrelated pairs, and HLA-unmatched unrelated pairs). Furthermore, HLA-matched sibling pairs with a female-to-male sex mismatch had a significantly higher overall skin GVHR grade and a higher ratio of high- versus low-grade skin GVHR than the sibling pairs with all other sex combinations. Patient age was not reflected in the skin explant result. In conclusion, the predictive value of the skin explant model for aGVHD varies depending on the clinical transplant protocols, such as the type of GVHD prophylaxis used. Nevertheless, the skin explant model remains a unique in vitro system that provides an in situ histopathologic readout for studying alloreactivity and human GVHD. The model has also the potential to aid the development of novel prophylaxis and treatment for GVHD.