For juvenile justice practitioners working with youth who have sexually offended, an accurate risk assessment instrument can help guide important decisions about placement, supervision, and treatment. However, current knowledge and practice in assessing the risk of sexual recidivism for youth is limited and there are few existing tools that are empirically valid and reliable. The project examined current practice and policy in the assessment, treatment, and management of juveniles with a history of sexual offending across multiple jurisdictions (Florida, New York, Oregon, Pennsylvania, and Virginia) and developed a prototype assessment tool, state-specific risk assessment models, and practical guidance for building a risk assessment for sexual recidivism in juvenile justice settings. Key findings from the project highlight that predicting sexual recidivism among youth entails numerous inherent challenges due to the low frequency of occurrence, chief among which is the lack of reliability in risk prediction. Simulation analysis reveals that a risk prediction model that performs adequately in one setting often reversely classifies individuals in another setting while adopting an off-the-shelf assessment tool should be avoided due to extensive customization and testing requirements. A prototype risk assessment tool was developed out of a multi-state model due to features more widely applicable than a jurisdiction-specific models. The prototype developed, an example of how machine learning algorithms applications of risk assessment can be implemented in practice, provides a platform for improving current practice in sex offense risk assessment with advanced technology and existing administrative data.