In the past two decades restorative justice (RJ) has been the subject of more rigorous criminological research than perhaps any other strategy for crime prevention and victim support. A misalignment between practice and research, however, has resulted in much confusion about what practices are, or are not, supported by the existing research base. This confusion raises the moral problem of doing things to people without evidence those things do no harm. In what has become a wide array of justice practices called ‘restorative,’ there are serious risks of both direct and indirect harm in promoting--or even condoning-- untested practices: 1) Many practices remain untested, despite claims that tests of some RJ practices support all RJ practices, so that the untested practices may be causing harm directly; 2) Practices that have been rigorously tested and found to be effective are not widely used, while untested RJ practices have arguably caused harm indirectly by diverting resources from practices known to be effective; 3) Victims of violent crime are indirectly harmed by the diversion of RJ resources to property crime, where evidence shows RJ is less effective. We therefore assert a moral obligation for RJ practitioners to assure that their work does no harm by promoting rigorous evaluations of what they are doing, and encouraging investment in tested strategies for the kinds of victims and offenders with whom RJ is known to have the strongest effects.