Supporting language in mathematics classrooms requires both curriculum material that follows language-responsive design principles and teaching practices that enact these principles with high instructional quality. This paper presents the analytic framework L-TRU, which was developed to assess language-responsive teaching practices quantitatively. The L-TRU framework draws upon Schoenfeld’s teaching for robust understanding (TRU) framework by adapting its five dimensions to language-responsive classrooms: Mathematical Richness, Cognitive Demand, Equitable Access, Agency, and Use of Student Contributions. It is extended by two further dimensions, namely, Discursive Demand and Connecting Registers. The adapted and extended L-TRU rating scheme was applied to 41 video-recorded lessons of 26 teachers who all used the same language-responsive curriculum material on percentages. The qualitative insights gained from selected transcripts reveal that the dimensions indeed capture important distinctions in valid ways. The analysis of interrater reliability and correlations confirms that distinct dimensions are captured with reliability. The quantitative overview of the ratings of 497 episodes shows that in spite of the shared curriculum material, a large variety of instructional practices were enacted: Consistently high quality was found in the dimensions Cognitive Demand and Equitable Access and a medium quality in Connecting Registers. The dimensions Agency, Discursive Demand and Use of Contributions show the largest variance among teachers, with Discursive Demand separating most. These findings empirically substantiate an important research tool for quantitatively capturing teaching practices with respect to their general mathematics instruction quality and language-responsive quality.