Adoptive T-cell therapy using CD3/CD28 costimulation likely requires in vivo generation of antigen specificity. Because CD28 promotes T H 1/T C 1 (T1) or T H 2/T C 2 (T2) differentiation, costimulation may generate donor T1 or T2 cells capable of differentially mediating allogeneic graft-versus-tumor (GVT) effects and graft-versus-host disease (GVHD). Costimulation under T1 or T2 conditions indeed generated murine T H 1/T C 1 cells secreting interleukin-2/interferon-␥ (IL-2/ IFN-␥) or T H 2/T C 2 cells secreting IL-4/IL-5/IL-10. In vivo, allogeneic T1 cells expanded, maintained T1 secretion, and acquired allospecificity involving IFN-␥ and IL-5. In contrast, allogeneic T2 cells expanded less and maintained T2 secretion but did not develop significant allospecificity. Allogeneic, but not syngeneic, T1 cells mediated a GVT effect against host-type breast cancer cells, as median survival time (MST) increased from 25.6 ؎ 2.6 (tumor controls) to 69.2 ؎ 5.9 days (P < 1.2 ؋ 10 ؊9 ). This T1-associated GVT effect operated independently of fasL because T1 cells from gld mice mediated tumor-free survival. In contrast, allogeneic T2 cells mediated a modest, noncurative GVT effect (MST, 29 ؎ 1.3 days; P < .0019). T1 recipients had moderate GVHD (histologic score, 4 of 12) that contributed to lethality after bone marrow transplantation; in contrast, T2 recipients had minimal GVHD (histologic score, 1 of 12). CD3/CD28 costimulation, therefore, generates T1 or T2 populations with differential in vivo capacity for expansion to alloantigen, resulting in differential GVT effects and GVHD.