The use of the human tumor cloning assay as a predictor of clinical response of human tumors to drugs is predicated on the hypothesis that the in vivo response of a tumor to a drug can be correlated with the in vitro response of cells derived from the tumor. To test this hypothesis, we utilized a murine tumor model in which the in vivo and in vitro responses of a tumor can be accurately and reproducibly compared. Drug activity was assessed in P388 leukemia with the standard in vivo antitumor assay (i.p. tumor/i.p. drug administration) and an in vitro assay wherein the ascites tumor cells are removed from mice, treated with a drug, and directly cloned in soft agar to measure clonogenic capacity. The response of P388 cells to analogues within four separate classes of antitumor agents, anthracyclines, anthraquinones, platinum(II) coordination complexes, and phosphinogold(I) complexes was evaluated. The clonogenic assay failed to discriminate between highly active in vivo antitumor agents and analogues with only marginal in vivo efficacy (i.e., doxorubicin and daunorubicin versus rhodomycins A and B, ametantrone versus NSC 276740, cisplatin versus transplatin, [Au(dppe)2]Cl versus [Au(depe)2]PF6. Furthermore, the in vitro clonogenic assay failed to detect carboplatin which was a highly active agent in vivo. The basis for these discrepancies was explored by a more detailed comparison of doxorubicin and rhodomycin B. In vivo or in vitro drug exposure with subsequent measurement of cell kill by the in vitro clonogenic and in vivo tumorigenic assay demonstrated that the in vitro assay overestimated the cytotoxic potency of the drugs relative to the tumorigenic assay. Treatment of tumors in vivo with doxorubicin at doses below the maximally tolerated dose in mice resulted in multiple log cell kill as measured in vitro or in vivo, whereas rhodomycin B was cytotoxic only at dose levels exceeding its maximally tolerated dose. The results indicate that a subset of tumor stem cells capable of forming colonies in soft agar are significantly more sensitive to the cytotoxic effects of anthracyclines than are in vivo tumorigenic stem cells. Cytotoxic potency as measured by an in vitro soft agar clonogenic assay is not an accurate predictor of in vivo antitumor efficacy even in a model in which ascites tumor cells are directly exposed to i.p. drug. The in vitro cytotoxicity assay is useful only as a nonselective prescreen and must be used in combination with other indicators of tumor cell selectivity and dose-limiting organ toxicity.