In Reason's safety model, high-reliability healthcare organisations are characterised by multiple layers of defensive barriers in depth associated with increased levels of safety in the care delivery system. However, there is very little empirical evidence describing and defining defensive barriers in healthcare settings or systematic analysis documenting the nature of breaches in these barriers. This study uses in situ simulation to identify defensive barriers and classify the nature of active and latent breaches in these barriers. An in situ simulation methodology was used to study team performance during obstetrics emergencies. The authors conducted 46 trials of in situ simulated obstetrics emergencies in two phases at six different hospitals involving 823 physicians, nurses and support staff from January 2006 to February 2008. These six hospitals included a university teaching hospital, two suburban community hospitals and three rural hospitals. The authors created a high-fidelity simulation by developing scenarios based on actual sentinel events. A total of 965 breaches were identified by participants in 46 simulation trials. Of the 965 breaches, 461 (47.8%) were classified as latent conditions, and 494 (51.2%) were classified as active failures. In Reason's model, all sentinel events involve a breached protective layer. Understanding how protective layers breakdown is the first step to ensure patient safety and establish a high reliability. These findings suggest where to invest resources to help achieve a high reliability. In situ simulation helps recognise and remedy both active failures and latent conditions before they combine to cause bad outcomes.