Studies using animal models have demonstrated that transduction of genes encoding different cytokines into tumor cells results in a local recruitment of inflammatory cells that in turn can inhibit tumor growth. This is often accompanied by tumor antigen priming of the host immune system, which becomes resistant to subsequent challenge by the parental, untransduced tumor. Gene-transduced tumor cells have therefore been widely used as vaccines, although in the therapeutic setting their antitumor efficacy was limited to a few animal models. On the basis of this rationale, clinical studies were initiated, results of which are evaluated in this review to identify the reasons for their limited efficacy. We point out problems generated by the use of autologous versus allogeneic gene-transduced vaccines, by the choice of the appropriate cytokine(s), and by patient selection. Results of these studies are also compared with those obtained by peptide-based vaccines in similar groups of patients. Altogether, we conclude that improvements can be made in the construction of gene-modified vaccines by (1) using tumor cells known to express molecularly defined antigens, (2) introducing, in addition to genes encoding cytokines, genes encoding T cell costimulatory molecules, (3) increasing the amount of cytokine released locally by irradiated cells, and (4) coadministering adjuvant cytokines (IL-2 and IL-12) systemically in order to expand the T cell pool activated by vaccines.