In animal tumor models the dose-response curve for cytotoxic agents, especially cyclophosphamide, may be steep, but the slope and shape of this curve depends not only on the drug used but on the schedule of drug administration, the specific tumor type, tumor cell kinetics, and tumor mass. It might be anticipated from these studies that the human tumors most sensitive to dose effects would be leukemia, lymphoma, small-cell carcinoma of the lung, and testicular tumors rather than the low growth fraction, relatively less responsive tumors such as breast cancer. However, the clinical evidence for a steep dose-response curve in any tumor type is limited. For breast cancer such evidence is largely retrospective or derived from uncontrolled trials. The data available from randomized trials makes it seem unlikely that small, or even moderate, reductions in drug dose for nontrivial reasons will compromise the survival of patients with either early or metastatic disease. In spite of promising data from small trials, there is, as yet, inadequate evidence to justify the use of very-high-dose therapy and autologous marrow transplant outside the setting of a well-designed clinical trial. The value of high-dose therapy, intensive dose rate, and cumulative drug dose should each be studied in randomized controlled trials.