A consecutive sample of 120 male juvenile offenders was selected retrospectively from the Gainesville State School (Texas) clinical files. Adjudicated offenders are evaluated at a centralized assessment facility by the Texas Youth Commission; those determined to need maximum security are remanded to Gainesville State School and similar facilities. Three advanced doctoral students in clinical psychology participated in the admission evaluations. These clinicians were trained specifically in structured interviews and were provided with ongoing supervision. The two screening instruments were used to predict total, violent, self-injurious, and nonviolent infractions in a treatment-oriented facility for delinquents. In predicting the overall number of infractions, the MMPI-A was superior to the PCL:SV. The PCL correlated moderately with violent infractions among African-Americans but not among Anglo-Americans or Hispanic-Americans, suggesting possible systematic biases related to ethnicity. The study concludes that clinicians cannot afford to extrapolate across different measures, sample characteristics, or types of problematic behavior. Also, prediction models must take into account more than isolated traits. 2 tables and 37 references