The ErbB network is dysregulated in many solid tumors. To exploit this, we have developed a chimeric Ag receptor (CAR) named T1E28z that targets several pathogenetically relevant ErbB dimers. T1E28z is coexpressed with a chimeric cytokine receptor named 4αβ (combination termed T4), enabling the selective expansion of engineered T cells using IL-4. Human T4(+) T cells exhibit antitumor activity against several ErbB(+) cancer types. However, ErbB receptors are also expressed in several healthy tissues, raising concerns about toxic potential. In this study, we have evaluated safety of T4 immunotherapy in vivo using a SCID beige mouse model. We show that the human T1E28z CAR efficiently recognizes mouse ErbB(+) cells, rendering this species suitable to evaluate preclinical toxicity. Administration of T4(+) T cells using the i.v. or intratumoral routes achieves partial tumor regression without clinical or histopathologic toxicity. In contrast, when delivered i.p., tumor reduction is accompanied by dose-dependent side effects. Toxicity mediated by T4(+) T cells results from target recognition in both tumor and healthy tissues, leading to release of both human (IL-2/IFN-γ) and murine (IL-6) cytokines. In extreme cases, outcome is lethal. Both toxicity and IL-6 release can be ameliorated by prior macrophage depletion, consistent with clinical data that implicate IL-6 in this pathogenic event. These data demonstrate that CAR-induced cytokine release syndrome can be modeled in mice that express target Ag in an appropriate distribution. Furthermore, our findings argue that ErbB-retargeted T cells can achieve therapeutic benefit in the absence of unacceptable toxicity, providing that route of administration and dose are carefully optimized.