Much of the research on students' understanding of the greenhouse effect and global warming reports poor results. Students are claimed to hold misconceptions and naive beliefs, and the impact of teaching on their conceptions is also low. In the present study, these results are called into question, and it is argued that they may to a large extent be seen as artifacts of the research methods deployed, in particular when written questionnaires are used. When following students' project work in school over a long period, many of the misunderstandings reported in the literature do not appear. It is argued that the appropriation and use of scientific language when discussing complex socioscientific issues is a gradual process. When observing the language and mediational means students use over time, it is obvious that they are able to identify and use central distinctions in their interactions. They are also able to make productive use of texts and other materials that allow them to successively approximate scientific modes of reasoning. Thus, what students know emerges in communicative practices where they interact with others and with cultural tools in a focused activity. It is argued that students' knowledge of complex multidisciplinary phenomena of this kind may be particularly ill-suited to conventional questionnaire types of testing.