In this study, a novel approach to antitumor therapy was devised by generating a chimeric tumor-targeted killer protein, referred to as immunocasp-3, that comprises a single-chain anti-erbB2/HER2 antibody with a NH₂-terminal signal sequence, a Pseudomonas exotoxin A translocation domain, and a constitutively active caspase-3 molecule. In principle, cells transfected with the immunocasp-3 gene would express and secrete the chimeric protein, which then binds to HER2-overexpressing tumor cells. Subsequent cleavage of the constitutively active capase-3 domain from the immunocasp-3 molecule and its release from internalized vesicles would lead to apoptotic tumor cell death. To test this strategy, we transduced human lymphoma Jurkat cells with a chimeric immunocasp-3 gene expression vector and showed that they not only expressed and secreted the fusion protein but also selectively killed tumor cells overexpressing HER2 in vitro. i.v. injection of the transduced Jurkat cells led to tumor regression in a mouse xenograft model because of continuous secretion of immunocasp-3 by the transduced cells. The growth of HER2-positive tumor cells in this model was inhibited by i.m. as well as intratumor injection of immunocasp-3 expression plasmid DNA, indicating that the immunocasp-3 molecules secreted by transfected cells have systematic antitumor activity. We conclude that the immunocasp-3 molecule, combining the properties of a tumor-specific antibody with the proapoptotic activity of a caspase, has potent and selective antitumor activity, either as cell-based therapy or as a DNA vaccine. These findings provide a compelling rationale for therapeutic protocols designed for erbB2/HER2-positive tumors.