The purpose of this work is to analyze the strategy used for the discovery of potentially clinically effective drugs by means of transplantable tumor systems in mice. With the present strategy used by the National Cancer Institute and the European Organization for Research on Treatment of Cancer, most new compounds are initially tested in the P388 leukemia model. Drugs showing activity in this prescreen are subsequently evaluated in a panel of eight screens (B16, CD8, Co38, L1210, LL, CX-1, LX-1, and MX-1). Positivity in any one of these screens qualifies a new drug for further preclinical studies leading to phase I-II clinical trial. Our analyses of the experimental screening data for 1949 compounds, and the comparison of results in humans and mice for 69 drugs that have been clinically evaluated against solid tumors, have indicated that the correlation of screening and clinical results is low and that a modified approach using four systems (P388, B16, L1210, and MX-1) would have been sufficient to uncover most of the drugs presently shown to be clinically active. In order to take advantage also of the information provided by other screens of the panel, a three-stage strategy can be used. In the first stage, compounds would be tested in the P388 prescreen, with a slightly higher level of response being required (T/C% greater than or equal to 125 instead of 120). In the second stage, prescreen-positive drugs would be tested in the B16, L1210, and MX-1 systems. Drugs positive in any one of these screens may then be tested in Co38, LL, and CX-1 to determine priorities for clinical study. For new chemical structures of particular promise, this third stage could be dropped. LX-1 and CD8 are not useful in this context. Such a strategy would be much more economical and faster than the automatic use of nine systems, and the resultant savings should permit the discovery of an increased number of potentially active compounds.