Three points are made in this paper. It is found that concurrent controlled randomized trials are most useful (especially Phase II trials) in speeding up the introduction of better treatments for cancer patients. A scheme of continuing comparative trials, proposed by Theodore Colton, which require no statistical significance tests, is strongly recommended. The "historical-controls" approach of Gehan and Freireich is examined and is largely rejected because it appears to be unworkable. It is urged that patient characteristics that are associated with good response be determined, for the process of giving each patient the optimal treatment for him. This is referred to as "the epidemiology of treatment." Two examples of how early uncontrolled results are misleading in evaluating potential cancer treatments are given.