Adoptive T-cell therapy with gene-modified T cells expressing a tumor-reactive T-cell receptor or chimeric antigen receptor (CAR) is a rapidly growing field of translational medicine and has shown success in the treatment of B-cell malignancies and solid tumors. In all reported trials, patients have received T-cell products comprising random compositions of CD4(+) and CD8(+) naive and memory T cells, meaning that each patient received a different therapeutic agent. This variation may have influenced the efficacy of T-cell therapy, and complicates comparison of outcomes between different patients and across trials. We analyzed CD19 CAR-expressing effector T cells derived from different subsets (CD4(+)/CD8(+) naive, central memory, effector memory). T cells derived from each of the subsets were efficiently transduced and expanded, but showed clear differences in effector function and proliferation in vitro and in vivo. Combining the most potent CD4(+) and CD8(+) CAR-expressing subsets, resulted in synergistic antitumor effects in vivo. We show that CAR-T-cell products generated from defined T-cell subsets can provide uniform potency compared with products derived from unselected T cells that vary in phenotypic composition. These findings have important implications for the formulation of T-cell products for adoptive therapies.