The successful development of a soft agar clonogenic assay for human tumor stem cells provides an in vitro technique with a high degree of accuracy for predicting in vivo clinical response to standard anticancer drugs. We used this system to conduct an "in vitro phase II trial" in human ovarian cancer and melanoma. This approach can potentially identify active phase I--II drugs suitable for treatment of given tumor types for specific patients and eliminates the need to subject patients (who would be predicted not to respond) to toxic side effects. In vitro sensitivity for new agents was operationally defined as at least a 70% reduction of tumor colony-forming units (TCFU) at concentrations which are readily achievable pharmacologically. The new agents AMSA and vindesine (as well as vinblastine) appeared to have activity in melanoma, while PALA and thymidine were inactive. Pentamethylmelamine, mitomycin C, methyl-GAG, and AMSA were relatively ineffective in ovarian cancer. Vinblastine and vindesine had definite activity. The human tumor stem cell assay may thus provide the basis for a useful alternative to the current clinical phase II testing approach for identifying antitumor activity of new agents. Validation of this concept with correlative in vitro and in vivo phase II trials of new agents in patients with tumor types predicted to be sensitive is clearly warranted.