We have known for decades that IT projects often fail. The usual explanation is that the cause is poor project management, poor cost estimation, poor requirements, etc. But how can we prevent these causes? To answer this question, it is not sufcient to know that there was poor project management, etc. Would it sufce to educate better project managers? And what would we have to teach them that they don't learn today?We have to know what actually took place in the projects, identify the root causes, and nd ways to prevent them (cures). This is similar to accident investigations in aviation. This paper reports the results of ve accident investigations of large government IT projects in Denmark. We identied 37 root causes and 22 potential cures. Surprisingly, only one of the causes is programming-related. Each project suffered from around 15 of the causes. Twenty-seven of the causes are not reported in the research literature on IT project failures (e.g. surprises with system integration and wrong estimate of human performance). Half of the cures are familiar to developers, but were ignored in the specic project (example: usability test). The rest are unfamiliar (e.g. problem-oriented requirements and monitoring remaining work).