On the assumption that we are able to justify the institution of punishment, when may people be punished? More specifically, may a person be punished for an offence only after she has committed the offence? Christopher New has recently argued that, despite our intuitions to the contrary, there is in principle nothing which forbids 'punishment' before the offence has been committed, i.e. 'prepunishment' [2]. He claims that 'there may be room in our moral thought for the notion of prepunishment, ... it may only be epistemic, rather than moral, constraints that prevent us from practising it' ([2], p. 35). New presents a detailed example, about a person bent on over-speeding, but willing to pay a fine for the offence. This person is however willing to pay the fine only before over-speeding ([2], pp. 35-36). If it is certain beyond any reasonable doubt that the person will indeed commit the offence, but that then he could not be caught, what objection could we have to prepunishment? I explore the challenge presented by New, and argue that prepunishment is deeply ethically unacceptable. My argument, unlike New's intriguing suggestion, is not innovative: the problem with prepunishment derives, in the end, from the widely recognized need to respect persons and from the unacceptability of the 'punishment' of the innocent. New defends his suggestion against both consequentialist and retributivist objections, but we need not concern ourselves here with the former. I allow that prepunishment may be useful. We could well imagine a significant reduction in crime if selective prepunishment were applied to e.g. dangerous criminals with many previous convictions. But then consequentialists, and in particular utilitarians, have notoriously had difficulties in escaping the need to 'punish' the innocent (see [3]). The crucial question is whether someone who believes in (non-consequentialist) constraints of justice has the resources to reject New's suggestion. It should be emphasized that the existence of such constraints need not commit us to any strongly retributivist position. H. L. A. Hart's dual-level position, whereby the institution of punishment is justified by its utility, but in which the 'distribution' of punishment is constrained by the categorical unacceptability of 'punishing' those who have not committed an offence, could also, I claim, allow us to reject New's suggestion ([1], pp. 9f.). The crucial matter is clearly the deep value of taking care not to 'punish' the innocent. This value is not self-evident, but for our purposes putting it in doubt is not important. For New's strategy is to uphold this value but to