Risk assessment procedures occupy a central role in the criminal justice system decision making process and typically involve a prediction about the likelihood that an individual will re-offend (Bonta, 1996, 2002; Gottfredson, 1987; Gottfredson & Moriarty, 2006; Hoge, 2002; Taxman, Cropsey, Young, & Wexler, 2007; Taxman & Thanner, 2006). The efficiency of risk assessment tools is determined largely by the predictive accuracy of the underlying analytical techniques and any improvement in accuracy is likely to result in significant benefits for public safety and offender rehabilitation. While risk assessment tools have typically been developed using traditional statistical techniques such as regression models that involve testing relationships to provide evidence for or against a given hypothesis, advances in the development of statistical computation techniques in the Information Technology (IT) fields of Artificial Intelligence (AI) and Knowledge Discovery and Data Mining (KDD) have the potential to improve the predictive accuracy of risk assessments in criminal justice (Caulkins, Cohen, Gorr, & Wei, 1996; Palocsay, Wang, & Brookshire, 2000). The aim of the project was to apply techniques from the field of KDD (neural networks and decision trees) to criminal justice data to determine whether these techniques could be used to improve the predictive accuracy of models developed to predict risk of re-offending over base cases and commonly applied statistical methods. To accomplish this aim, the predictive accuracy of base cases, models developed using traditional statistical techniques, and models developed using neural networks and decision trees that predicted juvenile re-offending were compared.