We address the problem of selecting a treatment for phase II evaluation when several candidate treatments emerging from phase I testing are available. A pre-phase II Bayesian selection design which randomizes patients among treatments is proposed. The patient group in the trial has prognosis more favourable than that of phase I patients but less favourable than the target group of the subsequent phase II trial. The patient response rate distribution in each treatment arm is continually updated during the trial for comparison with early termination cutoffs, and the best final treatment must satisfy a minimal posterior efficacy criterion. The primary aim is to replace the usual informal treatment selection process with a fair comparison formally based on a combination of prior opinion and clinical data.