The present study investigates the differential effects of cooperative learning with or without metacognitive instruction on making sense of graphs. Participants were 196 eighth-graders who studied in six classrooms. Data were analyzed by using quantitative and qualitative methods. Results indicated that students who were exposed to the metacognitive instruction within cooperative learning (COOP+META) significantly outperformed their counterparts who were exposed to cooperative learning with no metacognitive instruction (COOP). The positive effects of COOP+META were observed on both graph interpretation and graph construction (transfer task) with regard to alternative conceptions. Furthermore, observations indicated differential characteristics of discourse behaviors during small group interaction under these methods. The practical implications of the study are discussed.