Recent clinical trials have shown that evasion of CD8⁺ T-cell responses against viral capsid is critical for successful liver-directed gene therapy with adeno-associated viral (AAV) vectors for hemophilia. Preclinical models to test whether use of alternate serotypes or capsid variants could avoid this deleterious response have been lacking. Here, the ability of CD8⁺ T cells (“cap-CD8,” specific for a capsid epitope presented by human B*0702 or murine H2-L^d molecules) to target AAV-infected hepatocytes was investigated. In a murine model based on adoptive transfer of ex vivo expanded cap-CD8, AAV2-transduced livers showed CD8⁺ T-cell infiltrates, transaminitis, significant reduction in factor IX transgene expression, and loss of transduced hepatocytes. AAV8 gene transfer resulted in prolonged susceptibility to cap-CD8, consistent with recent clinical findings. In contrast, using an AAV2(Y-F) mutant capsid, which is known to be less degraded by proteasomes, preserved transgene expression and largely avoided hepatotoxicity. In vitro assays confirmed reduced major histocompatibility complex class I presentation of this capsid and killing of human or murine hepatocytes compared with AAV2. In conclusion, AAV capsids can be engineered to substantially reduce the risk of destruction by cytotoxic T lymphocytes, whereas use of alternative serotypes per se does not circumvent this obstacle.