This paper shows that escalating fines emerge in a generalized version of the canonical Becker (1968) model if the authority (i) conditions optimal fines on offender histories, and (ii) does not fully credit offender gains to social welfare. We demonstrate that escalation is driven by decreasing fines for first-time offenders rather than increasing fines for repeat offenders. The authority can gain from lowering the fine for first-time offenders with a clean history, thereby redistributing additional offender gains to society. In contrast, the authority cannot gain from increasing the fine for repeat offenders because of their positive selection (Tirole 2016). Our analysis nests optimal law enforcement with uncertain detection and behavior-based monopoly pricing with imperfect customer recognition.