The most serious problem in current gene therapy is discrepancies between experimental data and actual clinical outcomes, which may be due to insufficient analyses and/or inappropriate animal models. We have explored suicide gene therapy by using various clinically relevant animal models and doubt the clinical use of maximal suicide gene expression, which has been generally recommended. To explore this subject further, we studied what expression level of suicide gene and what promoter led to the maximal clinical benefit in the case of hepatic metastatic cancer in mice. Therapeutic and adverse side effects of 4 adenoviral vectors that express herpes simplex virus thymidine kinase (HSV-tk) under different promoters were scrupulously investigated in 2 mouse models of hepatic metastasis of gastric cancer that possess clinical characteristics. Surprisingly, increases in HSV-tk expression beyond a certain point, achieved by the Rous sarcoma virus long terminal repeat promoter, not only enhanced the adverse side effects of lethal hepatotoxicity and ganciclovir-independent cytotoxicity but also failed to further increase therapeutic potential. Moreover, the carcinoembryonic antigen (CEA) tumor-specific promoter, the therapeutic potential of which had been underestimated, was much more useful-even in the case of low CEA-producing cancer-than had been previously reported. In conclusion, the optimal therapeutic expression level of a suicide gene is a novel concept and a crucial factor for successful cancer gene therapy. The present results, which contradict those of previous studies, alert researchers about possible problems with ongoing and future clinical trials that lack this concept.