Randomized trials are usually too small to provide decisive evidence on the effectiveness of cancer therapy, and most therapeutic developments have been based on studies of tumour response. This approach, together with survival comparisons using historical controls, has led to the development of effective chemotherapy for a few very responsive tumours, but not for most common cancers, where the impact of treatment on survival is quite small. It is difficult to distinguish effective and ineffective innovations in the treatment of such cancers, and consistent progress, in which the more effective treatments are developed and the less effective ones discarded, cannot be achieved without comparisons based on large numbers of randomized patients. The combined evidence on adjuvant therapy for breast cancer and colorectal cancer is reviewed, and a new meta-analysis of published chemotherapy trials in advanced ovarian cancer is presented. The evidence that certain adjuvant regimens can prolong survival is conclusive for breast cancer and quite strong for colorectal cancer, and there is suggestive evidence that cis-platinum prolongs survival in advanced ovarian cancer. Such results illustrate the importance of large randomized trials in evaluating and improving the treatment of common cancers. The major difficulty in achieving large patient entry to randomized studies is the reluctance of many oncologists to participate in collaborative clinical research. The reasons underlying this reluctance, and the way in which such collaboration should be organized to meet these objections, are also discussed.