SUMMARY An adaptive procedure is proposed for a comparative clinical trial of two treatments in which the response variable has a normal distribution. Patients are assumed to 'arrive' in twos, and at each stage, after each pair has been treated, a decision is made either to continue to the next stage, or to suspend randomization and to assign patients to the 'best' treatment. This leads to stopping rules which are tabulated for a variety of patient horizons. In general, trials will be relatively short, and the patient horizon is less influential in determining trial duration than has been suggested by other authors.