While collaborative learning is prevailing in schools, most of its research focuses mainly on small group collaborative processes, in one shot practice. In order to investigate the nature and the mechanism of the collaborative learning in a larger group, this paper presents analyses of a classroom in which 21 third-graders collaboratively discussed predictions about the results of a series of experiments as a whole class over twelve course hours and became able to grasp a rudimentary scientific concept of the atomic theory. We analyzed their levels of achievement, conversation patterns, selection sequences of predictions, and the contents of the utterances. The results revealed that all of the children succeeded in expressing their grasp of a rudimentary scientific concept, yet their routes to the achievement were diverse. A preliminary qualitative analysis of two children’s utterances shows that while their models were similar during the first half of the course, their differences became more explicit toward the end, which resulted in intense discussion between the two in front of the other children. The diversity observed in the entire course of this class and emphasized toward the end by these two children’s explicit, focused dialogue could have contributed to the successful conceptual change for the entire class members.