We examine the effect of educational attainment on criminal behavior using random assignment into Job Corps (JC) – the United States' largest education and vocational training program for disadvantaged youth – as a source of exogenous variability in educational attainment. We allow such random assignment to violate the exclusion restriction when used as an instrument by employing nonparametric bounds. The attainment of a degree is estimated to reduce arrest rates by at most 11.8 percentage points (about 32.6%). We also find suggestive evidence that the effects may be larger for males relative to females, and larger for black males relative to white males. Remarkably, our 95 percent confidence intervals on the causal effect of education on arrests are very similar to the corresponding confidence intervals on the same effect from studies exploiting changes in compulsory schooling laws as an instrumental variable in the estimation of the effect of education on arrest rates (e.g., Lochner and Moretti, 2004).