OBJECTIVE: To determine, in a community-based sample of slow-to-talk toddlers, the extent to which specific maternal responsive behaviors at 24 months predict child language at 24 and 36 months. METHODS: Mother-child dyads were recruited for this prospective longitudinal study from 3 local government areas spanning low, middle, and high socioeconomic status in Melbourne, Australia. At child age 18 months, 1138 parents completed a 100-word expressive vocabulary checklist; the 251 (22.1%) children scoring ≤20th percentile were then followed up to comprise this study. PREDICTORS: Six maternal responsive behaviors (imitations, interpretations, labels, expansions, supportive directives and responsive questions) were derived from mother-child free-play videos collected at 24 months of age and coded using the Observer XT system. OUTCOMES: Expressive and receptive language measured at 24 and 36 months of age (Preschool Language Scale-4), blind to maternal responsiveness ratings. RESULTS: Two hundred and twenty-six of the 251 (90.0%) mother-child dyads were followed up at 36 months. In confounder-adjusted linear regression analyses, expansions, imitations, and responsive questions were strongly associated with better receptive and expressive language at 24 and 36 months. Labels unexpectedly predicted poorer expressive language at 36 months. Expansions were the only maternal behavior that predicted improvement in language between 24 and 36 months. CONCLUSIONS: Maternal responsive behaviors, particularly expansions, offer promise in enhancing early language learning in slow-to-talk toddlers. Parent-child interactions characterized by frequent use of maternal labels at 24 months could also be a predictive marker of those slow-to-talk toddlers at greater risk of persistent language problems.