|Abstract:|| A number of features of surgical interventions make their evaluation in randomised trials more difficult. For example, new procedures may be technically difficult to deliver so that surgical performance takes time to stabilise and outcomes may vary over time and between different surgeons. Moreover, there is a lack of trials culture among surgeons, so that efficient designs that are acceptable to the community are required. This talk will review some of the practical and technical issues involved in evaluations of surgical interventions, focusing on our work on the use of routine databases to explore learning curves, clustering between surgeons and the multiple component nature of the interventions. Some implications for design and analysis will be discussed and the limitations will be illustrated through the AMAZE trial in cardiac surgery patients who have a history of atrial fibrillation (irregular heart rhythm).