Economic theories of deterrence have primarily been built on static models. A common and serious shortcoming of the existing dynamic deterrence models is the assumption of a twoperiod structure that ignores recidivism. The aims of this paper are to formulate and solve a general dynamic deterrence model that incorporates recidivistic behaviour, to explore its implications, and to derive some testable predictions. The analysis shows how the value and the intensity of engaging in illegal activity change over time, highlights the weaknesses of twoperiod deterrence models and compares the deterrent effectiveness of increasing the likelihood of punishment versus the severity of punishment. Finally, the recidivistic model provides a structural foundation for the widely used stochastic-process models of crime in operations research and criminology.