Although previous meta-analyses have documented the efficacy of computer-assisted statistics instruction, the current study examined a range of specific features that presumably influence its effectiveness, such as the level of learner engagement, learner control, and the nature of feedback. In 45 experimental studies with a control condition, computer-assisted statistics instruction provided a meaningful average performance advantage (d = 0.33). Because of great methodological heterogeneity among the studies, the authors employed a conservative but appropriate mixed effects model to examine potential moderator effects. The authors' analyses revealed three statistically significant findings. Larger effects were reported in studies in which treatment groups received more instructional time than control groups, in studies that recruited graduate students as participants, and in studies employing an embedded assessment. A newly developed second order standardized mean effect size, d[subscript diff], reveals that additional study characteristics may serve as meaningful moderators. Tight experimental control is needed to assess the importance of specific instructional features in computer-assisted statistics instruction. (Contains 3 tables and 1 figure.)