Externalizing behavior symptoms (EBS) in childhood are a strong predictor of future conduct problems. This study evaluated their predictive accuracy using logistic regression and receiver operating characteristic curve techniques. EBS, alone and in combination with other child and familial risk factors, were used to predict conduct problems 30 months later in a nonclinic population of kindergartners and Grade 1 children. The sensitivity (Sn) and positive predictive value (PPV) of EBS alone were below preset criteria of > or = 50% for each (prevalence < or = 15%). Sn and PPV increased when other child and familial factors were combined with symptoms but did not exceed the preset criteria. From a developmental perspective, substantial stability of EBS exists over time. However, from the perspective of prevention science, significant levels of misclassification will occur when EBS are used to designate high-risk status under the low-prevalence conditions of normal populations.