This paper develops and empirically examines a dynamic model of decisions to work, invest in human capital, and commit crime. By making all three activities endogenous, the model makes a number of new and interesting contributions to the study of crime. First, the model explains why older, more intelligent, and more educated workers tend to commit less of some property crimes than others. Age and education are more negatively correlated with crimes requiring little skill. Second, the model is useful for analyzing the impacts of education, training, and work subsidies on criminal behavior. It predicts that all three subsidy policies can reduce criminal activity. However, short-term wage subsidies only temporarily reduce crime, at the expense of increasing future crime rates. Third, unobserved age di erences in on-the-job skill investment explain why wages and crime are more negatively correlated at older ages: at later ages, wages more accurately re ect skill levels and the true opportunity cost of crime. Fourth, the model predicts a rise in youth crime should accompany the recent rise in returns to skill; however, adult crime rates may rise or fall since the most able are likely to reduce their criminal activity when older while the least able increase theirs. Finally, the model suggests that law enforcement policies increase education, training, and labor supply, while reducing criminal activity.A numberof testable implications of the model are empirically studied using data from the National Longitudinal Survey of Youth NLSY, Current P opulation Survey CPS, and Uniform Crime Reports UCR. Both ability and high school graduation are found to signi cantly reduce criminal participation among young men in the NLSY. High school graduation also reduces the probability that a young man will become incarcerated sometime in the following ve years. While the impact of high school graduation on criminal participation declines with age, its e ect on incarceration is large and relatively stable throughout young adulthood. We also estimate the deterrent e ect of more severe punishment, which appears to bestrong in the NLSY. Evidence from the UCR and CPS supports our individual-level ndings: states with higher high school graduation rates and more severe punishment policies have lower index property crime rates. A numberof other predictions are supported by the data, suggesting that the model is useful for studying the interactions of education, work, and crime.