We systematically examined relations among 6 measures of child language derived from 3 sources, including observations of the child's speech with mother, experimenter assessments, and maternal reports. A total of 184 20-month-olds and their mothers contributed complete information about child language comprehension and expression. Correlations of child language measures with socioeconomic status and maternal education were accounted for, as were correlations of child language measures with mothers' verbal intelligence, maternal report measures with mothers' tendency to respond in a socially desirable fashion, and experimenter assessments with child social competence. Structural equation modeling supported (1) strong relations among child language measures derived from observations of the child's speech with mother, experimenter assessments, and maternal reports; (2) the loading of multiple measures of child language from different sources on a single latent construct of vocabulary competence; and (3) the predictive validity of the vocabulary competence latent variable at 20 months, as well as receptive vocabulary specifically, for both verbal and performance IQ (verbal better than performance) at 48 months. Neither an index of child monologing (a nonvocabulary language measure) nor symbolic play (a nonlinguistic representational measure) covaried with vocabulary competence. Girls consistently outperformed boys on individual language measures, but no differences emerged in any model in the fit for boys and girls.