A positive association between compliance and clinical outcome has been observed in several randomized, controlled, clinical trials. This association, seen in the placebo-treated group as well as the active-treatment group, clarifies the possibility that data analyses incorporating estimates of protocol adherence are potentially biased. In the presence of non-compliance, or missing data from any cause, several statistical analyses may seem plausible, with none clearly superior to the others. These may include an analysis of all patients randomized, with imputed values for missing data, and an analysis restricted to protocol-adherent patients. The recommended approach is a conservative one that examines consistency among the plausible analyses. Using compliance data in trial conduct can also introduce bias into trial results by inducing differential treatment of compliers and non-compliers. This possibility arises, for instance, when adherence is affected by the randomized treatment. Non-compliance can have a substantial impact on statistical power and sample size requirements in a clinical trial. Under certain assumptions, required sample sizes are doubled with 30% non-compliance and tripled with 40% non-compliance.