A procedure was described for constructing, empirically, a classification system for children’s disturbed behavior. The procedure involved three steps: collection of referral and observation data int he clinic setting, factor analysis of this matrix, and analysis of factor profiles to determine homogeneous classes. Data were collected for 100 boys referred for diagnosis to 4 child agencies. 149 observation items and the presence or absence of 38 symptoms were subjectedto the Wherry-Gaylord procedure for extracting factors. Five factors resulted from this analysis: hyperactive, withdrawn, immature, aggressive, anxious. Analysis of the factor profiles showed that lcassifying on the basis of a single symptom, or the child’s highest factor scores, results in classes which are significantly less homogeneous than grouping according to similarities in rank, level, and scatter of the profile.