Recently there has been increased interest among researchers and those working in the Juvenile Justice System to develop risk assessment instruments (RAIs) to predict violence among juveniles. Although there are violence prediction instruments designed for use with adults that have been widely studied, few such instruments exist that have been specifically designed for use with juvenile populations, and none have been sufficiently evaluated for predictive accuracy. The purpose of this study was to evaluate whether a RAI could be constructed to accurately predict a specific type of recidivism, violent recidivism, among juvenile offenders. Two RAIs were created using separate methods of instrument construction, the Burgess method and predictive attribute analysis. Both instruments were created and tested for predictive accuracy using a sub-sample from the Pittsburgh Youth Study. Results showed that both RAIs were able to predict violent recidivism to a degree, however, neither instrument performed well when tested on the validation sample. The limitations of commonly used measures of predictive accuracy are discussed in relation to the findings reported here and in other validation studies with similar instruments. Recommendations for constructing RAIs to predict violence among juveniles are outlined, and some cautionary remarks are made with regard to their use.