This paper argues that current safety assessments methodologies and models must be revised in order to reflect what is now known about the complexity of technological disasters. It first introduces the normal accident debate and indicates three types of issues that maintain the quarrel unsettled: the retrospective fallacy, the unanticipated consequences of technological and social actions and the difficulty to establish normative frames. Second, it argues, despite the 'normal accident' background, that there is much room for improvement in the way safety is assessed, if one considers closely, for instance, the lessons from technological disasters. The Macondo well case study illustrates the complexity of the problem. As safety assessments rely for a good part on safety management system audits (although the situation differs according to the high risk industries concerned), it is found relevant to start questioning their limits. This is what is explored in a third section, based on literature, empirical studies, official documents from high-risk companies and theory. The conclusion is that the audit rationale is not adapted to address the complexity of high-risk socio technical system and what we know from major accidents investigations. Fourth, some authors addressing the limits of current safety assessment practices are discussed, and ideas to move forward are introduced.
