OBJECTIVE: To create an obsessive-compulsive disorder subscale (OCS) of the Child Behavior Checklist (CBCL) and to determine its internal consistency, sensitivity, specificity, and positive and negative predictive power to identify obsessive-compulsive disorder (OCD) in children and adolescents. METHODS: Three samples of equal size (n = 73) of children and adolescents, matched for age, gender, and race, were selected for these analyses: 1) a clinically ascertained OCD group, 2) a psychiatrically treated group whose records revealed no evidence of OCD, and 3) a general population control group. An OCS was created by applying factor analysis to 11 CBCL items. Examinations of internal consistency, sensitivity, specificity, and positive and negative predictive value were undertaken. RESULTS: Of 11 items hypothesized to predict OCD, 8 items were retained after factor analyses (smallest factor loading: 0.49) and used to calculate OCS scores. The retained items displayed excellent internal consistency (Cronbach's alpha coefficient = 0.84). OCD participants had significantly higher OCS scores than either psychiatrically treated or general population control groups. With the use of the 2 cutoff scores closest to the true rate of OCD in the overall sample, sensitivity was 75.3% to 84.9%, specificity was 82.2% to 92.5%, positive predictive value was 70.5% to 83.3%, and negative predictive value was 88.2% to 91.6%. CONCLUSION: The performance of the proposed CBCL OCS compares favorably with that of the only previously studied screening instrument for OCD, the Leyton Obsessional Inventory-Child Version. Unlike the Leyton Obsessional Inventory-Child Version, the CBCL is already in widespread use as a screen for most other forms of psychopathology. As the performance of the CBCL OCS will need to be replicated in other sample populations, data with various cutoff levels are provided to enable investigators and clinicians to tailor its use to specific study populations.