Most prison systems use quantitative instruments to classify and assign inmates to prison security levels commensurate to their level of risk. Bench and Allen (The Prison Journal 83(4):367-382, 2003) offer evidence that the assignment to higher security prisons produces elevated levels of misconduct independent of the individual’s propensity to commit misconduct. Chen and Shapiro (American Law and Economics Review, 2007) demonstrate that assignment to higher security level among inmates with the same classification scores increases post-release recidivism. Underlying both of these claims is the idea that the prison social environment is criminogenic. In this paper we examine the theoretical premises for this claim and present data from the only experiment that has been conducted that randomly assigns inmates to prison security levels and evaluates both prison misconduct and post-release recidivism. The experiment’s results show that inmates with a level III security classification who were randomly assigned to a security level III prison in the California prison system had a hazard rate of returning to prison that was 31% higher than that of their randomly selected counterparts who were assigned to a level I prison. Thus, the offenders’ classification assignments at admission determined their likelihood of returning to prison. There were no differences in the institutional serious misconduct rates of these same prisoners. These results are contradictory to a specific deterrence prediction and more consistent with peer influence and environmental strain theories. These results also raise important policy implications that challenge the way correctional administrators will have to think about the costs and benefits of separating inmates into homogeneous pools based on classification scores.