Latent profile analysis (LPA) was used to derive homogeneous subgroups within the Child/Adolescent Anxiety Multimodal Study sample (N = 488; 7-17 years, M = 10.69, SD = 2.80) and examine whether class membership predicted or moderated treatment response. Subgroups were identified on baseline multi-informant measures of variables most consistently associated with outcome (youth anxiety/diagnosis, impairment, family psychopathology/functioning). Subgroup membership was examined as a predictor/moderator of outcome across the four treatment conditions (CBT, Sertraline, CBT+Sertraline, pill placebo) at posttreatment (12 weeks) and open-extension follow-up (24 weeks). Four subgroups emerged: mild symptoms/impairment, moderate symptoms/impairment, moderate symptoms/impairment with family dysfunction/parental psychopathology, and severe symptoms/impairment. There were significant between-class differences on socioeconomic status (SES; lower reported SES in the moderate with family dysfunction/parental psychopathology class compared to the mild and moderate class) and age (older age in the severe symptoms class compared to the other three classes). Youth in the mild symptoms/impairment class showed lower posttreatment anxiety across conditions but reported significantly lower symptom severity at baseline. Controlling for demographic differences, response to treatment type did not differ across classes. Analyses indicate that elevated family dysfunction/parental psychopathology clusters primarily within one subgroup of anxious youth rather than mapping onto symptom severity, highlighting the utility of LPA for clarifying within-person combinations of predictor/moderator variables. Implications for development of interventions targeting class-relevant variables are discussed.