This study explores how student-generated questions can support argumentation in science. Students were asked to discuss which of two graphs showing the change in temperature with time when ice is heated to steam was correct. Four classes of students, aged 12-14 years, from two countries, first wrote questions about the phenomenon. Then, working in groups with members who differed in their views, they discussed possible answers. To help them structure their arguments, students were given a sheet with prompts to guide their thinking and another sheet on which to represent their argument diagrammatically. One group of students from each class was audiotaped. Data from both students' written work and the taped oral discourse were then analyzed for types of questions asked, the content and function of their talk, and the quality of arguments elicited. To illustrate the dynamic interaction between students' questions and the evolution of their arguments, the discourse of one group is presented as a case study and comparative analyses made with the discourse from the other three groups. Emerging from our analysis is a tentative explanatory model of how different forms of interaction and, in particular, questioning are needed for productive argumentation to occur. (Contains 6 tables and 6 figures.)