Several statistical designs for phase II studies have been proposed, but they are frequently misunderstood or not applied at all. In this review we describe the major characteristics of the available designs. To investigate the extent to which statistical designs were used in some recent phase II studies, and which designs were the most common, we did a survey of 145 trials involving treatment of breast cancer. Studies selected for the survey were published between 1995 and 1999 in one of seven specific oncology journals (all with impact factor consistently higher than 2). 94 of the studies (64.8%) did not have an identifiable statistical design. However, among the 51 studies with statistical design there was a notable heterogeneity in the type of design applied. We put together a list of factors associated with use of statistical design at univariate analysis. These factors included: referral to a previous phase I study, recent trial start date, private sponsorship, single-agent treatment, and multicentre organisation. Single-agent treatment (OR 2.35; 95% CI 1.01-5.51) and multicentre organisation (OR 3.24; 95% CI 1.47-7.15) were independently predictive of the presence of statistical design. Publication in journals with high impact factors and short intervals between the start of the study and publication were also correlated with statistical design.