Many claim that non-prosecution agreements (NPAs) reduce deterrence by mitigating the reputational sanctions that would otherwise be imposed on corporations through plea-bargains. They suggest, based on this claim, that NPAs ought to be used infrequently. This article presents a signalling model wherein reputational sanctions emerge as a result of noisy signals produced through a firm's prosecution. It is shown that, if, as claimed, NPAs provide third parties with less information regarding a firm's wrongdoings, then firms would be willing to pay an NPA premium to avoid convictions. Thus, the NPA premium can be chosen to induce only those firms which would otherwise be over-deterred to accept NPAs. Therefore, offering NPAs with high premia is superior to the option of not using NPAs. The article also characterizes optimal NPAs, and identifies relationships between deterrence; frequency of NPA use; firms’ characteristics; and NPA terms. It explains how these relationships can be exploited to form and test hypotheses on whether convictions obtained through plea-bargains cause greater reputational harm to firms than NPAs.