Disenchantment with the tort system and negligence standard in the United States is fueling interest in alternate compensation systems for medical injury. One possibility is experimentation with administrative "health courts," through which specialized adjudicators would utilize neutral experts to render compensability determinations. Compensation would be based not on negligence, but rather on a broader avoidable medical injury (avoidability) standard. Although considerable interest in health courts exists, stakeholders frequently express uncertainty about the meaning and operation of an avoidability standard. Three nations-Sweden, Denmark, and New Zealand-have long operated administrative schemes. We conducted interviews with administrators and stakeholders in these systems. Our goal was to garner lessons on how to operate a health court, and specifically, how to develop and apply alternate compensation criteria such as avoidability. This article reports our findings on the origins and operations of the systems, the evolution of their compensation criteria, and how these criteria are actually applied. We found that all three systems had their primary genesis in ensuring compensation for the injured, as opposed to sanctioning providers. All have abandoned the negligence standard. The Nordic systems use an avoidability standard, principally defined as injury that would not occur in the hands of the best practitioner. Their experience demonstrates that this definition is feasible to apply. New Zealand's recent move to a no-fault system sheds light on the benefits and drawbacks of a variety of compensation standards. Key lessons for successfully applying an alternate standard, such as avoidability, include a strict adherence to national precedent, the use of neutral and experienced experts, and a block on routine transfer of information from compensation investigations to disciplinary authorities. Importantly, all three nations are harnessing their systems' power to improve patient safety, and the avoidability standard appears to be well suited for this task.