A randomised controlled trial (RCT) is considered the hierarchical peak of evidence-based medicine and a general demand for any result to be evaluated by RCTs has evolved. Yet, many advances in operative surgery do not result from RCTs and many controversies remain without an RCT being conducted. A randomised comparison of laparoscopic versus open liver resection has recently been called for. Using such a trial and others as examples, we examine the limitations of randomised design in skill-dependant interventions. Surgical procedures are skill-dependant, constantly developing, irreversible and traumatising. Additionally, placebo control is usually unethical and adequate blinding difficult or impossible to accomplish. Under these circumstances, surgeon and patient participation will be problematic and the resulting data will tend to have low external validity. While some of these obstacles can be modified, others will remain. Nonrandomised, prospective cohort comparison has other weaknesses, but may add complementary data with good external validity. An alternative hierarchy of evidence is warranted in this field.