The paper develops a dynamic theoretical model and presents empirical evidence about the relationship between violence and education investments. Although some papers have estimated regressions to link educational outcomes and violence, no formal models have been developed yet. A theoretical model is crucial to understand the different channels through which violence affects education. Three channels are identified. First, violence can affect directly the utility of households and, therefore, it may modify the consumption of education. Second, extreme violence can destroy physical capital and create uncertainty, which will lower investment and production. In the long run, destruction of physical assets and drop in investment impact the income of households who in turn must reduce consumption and cutback investments in education. Third, violence can modify the rates of return of education, and therefore, can change the investment on education. We find violence indeed exerts a toll on education. School enrollment is less in Colombian municipalities with homicide rates above the national median. Moreover, the likelihood of school enrollment decreases as homicide rates rise for all group ages. The impact of homicide rates is larger than transferences from the national government to the local government earmarked for investment in education and health.