The primary purposes of a phase II trial are to identify promising treatment regimens for further testing in a subsequent phase III trial and to screen out unpromising regimens which do not warrant further testing. Unfortunately, for many phase III cancer trials the new therapies tested have not led to improvements over standard (control) therapy, despite promising results in phase II trials. Since there are an increasingly large number of new cancer agents and combinations of agents that need to be assessed in phase II trials, it is increasingly important to design such trials with care. This talk addresses some reasons that phase II trials may fail to predict success in phase III trials. These reasons focus on differences between phase II trials and phase III trials in terms of sample size, patients studied, endpoints, treatment regimens, control groups, and toxicity assessments.