Maternal responsiveness is important in early language development, but its measurement by direct observation has to date been unrealistic in community settings because of training and time constraints. We report on the development and cross-sectional comparison of a global rating of maternal responsiveness against a detailed measure of responsiveness at age 24 months.The community-based sample comprised 246 toddlers and their mothers, identified as being slow-to-talk at age 18 months within the Let's Learn Language population-based randomized controlled trial. At age 24 months, mother-child dyads were videotaped during 15 minutes of free-play and children undertook a standardized language assessment. Videos were blindly rated on both the new global measure, comprising a single rating of responsiveness on a five-point Likert scale, and a detailed rating of responsiveness known to predict language outcomes, comprising a sum of specific maternal responsive behaviours.The global rating scale required relatively little training and ratings could be conducted in real time. The global and detailed ratings of maternal responsiveness showed moderate correlation (r = 0.44; P < 0.001). Small positive correlations were found between the global rating and expressive (r = 0.23; P < 0.001), receptive (r = 0.28, P < 0.001) and total language (r = 0.28; P < 0.001) at age 24 months.The global rating scale was efficient and moderately effective as a measure of maternal responsiveness. It is possible that, combined with other risk measures including concurrent language skills, it could strengthen prediction of which children will and will not go on to experience lasting language difficulties.