This article describes the impact of learning in asynchronous discussion groups on students' levels of knowledge construction. A design-based approach enabled the comparison of two successive cohorts of students (N = 223 and N = 286) participating in discussion groups for one semester. Multilevel analyses were applied to uncover the influence of student, group, and task variables on the one hand, and the specific impact of a particular form of scripting--namely the assignment of roles to group members--on the other. Results indicate that a large part of the overall variability in students' level of knowledge construction can be attributed to the discussion assignment. More intensive and active individual participation in the discussion groups and adopting a positive attitude towards the learning environment also positively relates to a higher level of student knowledge construction. Task characteristics--differences between the consecutive discussion themes--appeared to significantly affect levels of knowledge construction, although further analysis revealed that these differences largely disappeared after correcting for task complexity. Finally, comparisons between both cohorts revealed that the introduction of student roles led to significantly higher levels of knowledge construction. An effect size of 0.5 was detected.