AbstractAn independent Review Board has proposed as a target that the worst possible accident should give rise to no requirement for off-site evacuation. This paper addresses the problem of identifying the most important uncertainties, in current understanding of phenomena and design issues, bearing on compliance with the target. The problem is addressed using a simplified characterisation of accident initiation and progression, which is quantified as much as possible from the nearest analogous fission data, appropriately modifid to reflect fusion's characteristics, and from earlier fusion safety studies. The system chosen for analysis is a reactor cooled by pressurised water, but the effects of alternative arrangements are noted and commented on. Uncertainties are quantified and propagated by probabilistic methods. The identification and ranking of dominant contributors to overall uncertainty is performed by a statistical test. Although accidents initiated by external hazards, such as aircraft impact, were not explicitly analysed, it has proved possible to draw some conclusions about them also. Favourable conclusions are drawn relating to the likehohood that fusion reactor designs will comply with the target. It is stressed that this study is not an attempt to perform a probabilistic risk assessment of a fusion reactor: rather, by setting the uncertainties of fusion safety-relevant phenomena in a broad risk perspective, it indicates an agenda for design and research effort.