Learning outcomes from intelligent tutoring systems (ITSs) tend to be quite strong, usually in the neighborhood of one standard deviation. However, most ITS designers use the learning outcomes from expert human tutoring as the gold standard (i.e., two standard deviations). What can be done, with the current state of the art, to increase learning from an ITS? One method is to modify the learning situation by asking students to use the ITS in pairs. To enhance performance, we drew upon the beneficial effects of structured peer collaboration. The results suggest that the intervention was successful. Pairs of students solved more problems and requested fewer bottom-out hints than individuals. To test the possibility that the effect was due to the best partner in the group directing the problem solving, a nominal groups analysis was conducted. A nominal group is a statistical pairing of the non-interacting individuals' performance. The results from the nominal groups replicated the same pattern of results, but with a reduced magnitude. This suggests that the best member may have contributed to some of the overall success of the pair, but does not completely explain their performance.