We have directly compared the efficacy of two immunotherapeutic strategies for the treatment of cancer: "vaccination" of tumor-bearing mice with genetically modified dendritic cells (DCs), and vaccination with genetically modified tumor cells. Using several different preexisting tumor models that make use of B16F10 melanoma cells expressing a target tumor antigen (human melanoma-associated gene [MAGE]-1), we found that vaccination with bone marrow-derived DCs engineered to express MAGE-1 via adenoviral-mediated gene transfer led to a dramatic decrease in the number of metastases in a lung metastasis model, and led to prolonged survival and some long-term cures in a subcutaneous preexisting tumor model. In contrast, vaccination with granulocyte/macrophage colony-stimulating factor (GM-CSF)-transduced tumor cells, previously shown to induce potent antitumor immunity in standard tumor challenge assays, led to a decreased therapeutic effect in the metastasis model and no effect in the subcutaneous tumor model. Further engineering of DCs to express either GM-CSF, tumor necrosis factor alpha, or CD40 ligand via retroviral-mediated gene transfer, led to a significantly increased therapeutic effect in the subcutaneous tumor model. The immunological mechanism, as shown for GM-CSF-transduced DCs, involves MAGE-1-specific CD4(+) and CD8(+) T cells. Expression of GM-CSF by DCs led to enhanced cytotoxic T lymphocyte activity, potentially mediated by increased numbers of DCs in draining lymph nodes. Our results suggest that clinical studies involving the vaccination with genetically modified DCs may be warranted.