BACKGROUND:In a community sample of slow-to-talk toddlers, we aimed to (a) quantify how well maternal responsive behaviors at age 2 years predict language ability at age 4 and (b) examine whether maternal responsive behaviors more accurately predict low language status at age 4 than does expressive vocabulary measured at age 2 years. DESIGN OR METHODS:Prospective community-based longitudinal study. At child age 18 months, 1,138 parents completed a 100-word expressive vocabulary checklist within a population survey; 251 (22.1%) children scored ≤20th percentile and were eligible for the current study. Potential predictors at 2 years were (a) responsive language behaviors derived from videotaped parent-child free-play samples and (b) late-talker status. Outcomes were (a) Clinical Evaluation of Language Fundamentals-Preschool Second Edition receptive and expressive language standard score at 4 years and (b) low language status (standard score > 1.25 standard deviations below the mean on expressive or receptive language). RESULTS:Two hundred eight (82.9% of 251) participants were retained to age 4. In adjusted linear regression analyses, maternal expansions predicted higher receptive (p < 0.001, partial R2 = 6.5%) and expressive (p < 0.001, partial R2 = 7.7%), whereas labels predicted lower receptive (p = 0.01, partial R2 = 2.8%) and expressive (p = 0.007, partial R2 = 3.5%) language scores at 4. The logistic regression model containing only responsive behaviors achieved "fair" predictive ability of low language status at age 4 (area under curve [AUC] = 0.79), slightly better than the model containing only late-talker status (AUC = 0.74). This improved to "good" predictive ability with inclusion of other known risk factors (AUC = 0.82). CONCLUSION:A combination of short measures of different dimensions, such as parent responsive behaviors, in addition to a child's earlier language skills increases the ability to predict language outcomes at age 4 to a precision that is approaching clinical value. Research to further enhance predictive values should be a priority, enabling health professionals to identify which slow-to-talk toddlers most likely will or will not experience later poorer language.