Small-group, text-based discussions are a prominent and effective instructional practice, but the literature on the effects of different group composition methods (i.e., homogeneous vs. heterogeneous ability grouping) has been inconclusive with few direct comparisons of the two grouping methods. A yearlong classroom-based intervention was conducted to examine the ways in which group composition influenced students' discourse and comprehension. Fourth- and fifth-grade students (N = 62) were randomly assigned to either a homogeneous or heterogeneous ability small-group discussion. All students engaged in Quality Talk, a theoretically- and empirically-supported intervention using small-group discussion to promote high-level comprehension. Multilevel modeling revealed that, on average, students displayed positive, statistically and practically significant gains in both basic and high-level comprehension performance over the course of Quality Talk. Further, our findings indicated heterogeneous ability grouping was more beneficial than homogeneous ability grouping for high-level comprehension, on average, with low-ability students struggling more in homogeneous grouping. With respect to student discourse, additional quantitative and qualitative analyses revealed group composition differences in terms of the frequency, duration, and quality of student questions and responses, as well as the types of discourse low-ability students enacted in homogeneous groups. This study expands upon the extant literature and informs future research and practice on group composition methods. [This paper was published in "Contemporary Educational Psychology" v51 p336-355 2017.]