Objective: To compare the diagnostic efficiency of six index tests as predictors of juvenile bipolar disorder in two large outpatient samples, aged 5 to 10 and 11 to 17 years, gathered from 1997 to 2002. Method: DSM-IV diagnosis was based on a semistructured diagnostic interview (Schedule for Affective Disorders and Schizophrenia for School-Age Children) with the parent and youth sequentially, blind to scores on the index tests. Participants were 318 youths aged 5 to 10 (50% with bipolar diagnoses) and 324 youths aged 11 to 17 (41% with bipolar diagnoses). Areas under the curve (AUCs) from receiver operating characteristic analyses and multilevel likelihood ratios quantified test performance. Results: Parent report (AUCs from 0.78 to 0.84 in both age groups) outperformed teacher (AUCs 0.57 in the younger sample and 0.70 in the older sample) or adolescent measures (AUCs 0.67 [General Behavior Inventory] and 0.71 [Youth Self-Report]) at identifying bipolar disorders. Combining tests did not produce clinically meaningful classification improvement. Conclusions: Parent report was more useful than teacher report or adolescent self-report on the index tests studied. Results generally replicated across both age groups. Parent report on these instruments could facilitate differential diagnosis of bipolar disorder in youths aged 5 to 17 years, especially by decreasing the rate of false-positive diagnoses.