The need for treatment response predictive biomarkers is being increasingly recognized in children and adolescents with psychiatric disorders. Structural gray matter abnormalities as a predictor of treatment outcome in pediatric bipolar disorder have not been systematically investigated, especially early in the illness course. With a prospective longitudinal study design, the present study enrolled 52 bipolar adolescents with no history of treatment with mood stabilizers or a therapeutic dose of antipsychotic drugs and 31 healthy controls. Patients were randomly assigned to treatment with quetiapine or lithium after pretreatment data collection. A hierarchical cluster analysis was performed using pretreatment cortical thickness data that identified two discrete patient subgroups. Compared to healthy subjects, patients in subgroup 1 (n = 16) showed widespread greater cortical thickness mainly across heteromodal cortex but also involving some regions of unimodal cortex, while those in subgroup 2 (n = 36) showed regional cortical thinning mainly in superior temporal and superior parietal regions. Patients within subgroup 1 showed a significantly higher response rate to quetiapine than those in subgroup 2 (100% vs 53%). No statistically significant difference was found in lithium response rate between the patient subgroups (63% vs 53%). Pretreatment clinical ratings and neuropsychological data did not differ across subgroups. Our findings suggest the existence of distinct and clinically relevant subgroups of pediatric bipolar patients, as defined by pattern of cortical thickness. These groups appear to differentially respond to antipsychotic treatment-notably with greater cortical thickness relative to controls predicting better treatment response.