Double-unit cord blood transplantation (DCBT) appears to enhance engraftment despite sustained hematopoiesis usually being derived from a single unit. To investigate DCBT biology, in vitro and murine models were established using cells from 39 patient grafts. Mononuclear cells (MNCs) and CD34(+) cells from each unit alone and in DCB combination were assessed for colony-forming cell and cobblestone area-forming cell potential, and multilineage engraftment in NOD/SCID/IL2R-γ(null) mice. In DCB assays, the contribution of each unit was measured by quantitative short tandem repeat region analysis. There was no correlation between colony-forming cell (n = 10) or cobblestone area-forming cell (n = 9) numbers and clinical engraftment, and both units contributed to DCB cocultures. In MNC transplantations in NOD/SCID/IL2R-γ(null) mice, each unit engrafted alone, but MNC DCBT demonstrated single-unit dominance that correlated with clinical engraftment in 18 of 21 cases (86%, P < .001). In contrast, unit dominance and clinical correlation were lost with CD34(+) DCBT (n = 11). However, add-back of CD34(-) to CD34(+) cells (n = 20) restored single-unit dominance with the dominant unit correlating not with clinical engraftment but also with the origin of the CD34(-) cells in all experiments. Thus, unit dominance is an in vivo phenomenon probably associated with a graft-versus-graft immune interaction mediated by CD34(-) cells.