We analyze the effectiveness of working in pairs on the Conceptual Survey of Electricity and Magnetism test in a calculus-based introductory physics course. Students who collaborated with a peer showed significantly larger normalized gain on individual testing than those who did not collaborate. We did not find statistically significant differences between the performance of students who were given an opportunity to formulate their own response before the peer discussions, compared to those who were not. Peer collaboration also shows evidence for the co-construction of knowledge. Discussions with individual students show that students themselves value peer interaction. We discuss the effect of pairing students with different individual achievements.