The sources of variability influencing the results of phase II trials are reviewed. Randomized designs for phase II testing are presented and evaluated. Phase II designs with "standard therapy" control groups are not found to be broadly useful. Designs which randomize among new agents or schedules appear to be of value both scientifically and logistically where patient accrual is adequate. The rationale and advantages of this design are described. The concept of ranking and selection of new agents and schedules is presented as an alternative to testing the null hypothesis of therapeutic equivalence. Sample size calculations demonstrate potential advantages of this approach in appropriate situations. The conduct of pilot or "phase II" studies of combinations is also discussed and randomized designs with early stopping rules are proposed.