This article reports reliability and validity data for the Direct Observation Form (DOF) of the Child Behavior Checklist. Observational data were collected on two samples of boys aged 6-11 in classroom settings. Interobserver agreement was high: r = .92 for behavior problem score and r = .83 for on-task score. Generalizability, as measured by the one-way intraclass correlation, was .86 and .71 for behavior problem score and on-task score, respectively. In terms of validity, DOF scores correlated significantly and in the expected directions with teacher-reported problem behavior, school performance, and adaptive functioning. In addition, boys who had been referred by their teachers due to problem behavior obtained significantly higher behavior problem scores and significantly lower on-task scores than a matched sample of normal boys observed in the same classrooms.