Many methods have been developed for managing groups in active learning classes, but little research has been done on the effect of group structure itself. Results are presented for an active learning physics class in which half of the class was placed in academically homogeneous groups while the other half was in heterogeneous groups. Students were given the Conceptual Survey of Electricity and Magnetism as a pretest and posttest, and also filled out surveys on their experiences in their groups. The study was intended to be continued for three years, but was terminated halfway through the second semester as the evidence had become sufficiently compelling that placing half of the class in academically heterogeneous groups was placing them at a significant disadvantage. Student feedback, pretest, and posttest data indicated that low- and middle-performing students benefited the most from academically homogeneous groups. Results for the one full semester of the study and the rationale for discontinuing are presented.