Computer modeling, a new form of knowledge production implying new theoretical and experimental practices, has since the 1950s become an important tool in scientific investigation. This paper analyzes computer simulation in two cases of atmospheric transport modeling. First, it investigates pioneering efforts by the meteorologist Heinz Fortak in the 1960s to construct Gaussian models of local and regional air pollution transport. Second, it deals with the much more complex simulation efforts by a Norwegian group around Anton Eliassen in the 1970s and 1980s to model long distance transport and transformation of pollutants. In both cases, the simulation of air pollution phenomena implied a set of specific practices. The use of the computer as a tool required drastic simplifications; the collection and preparation of an enormous amount of input data; the adoption of control procedures to validate the models; and the execution of ““computer experiments”” in sensitivity studies. Because of uncertainty and ignorance, scientists enjoyed a large freedom of choice in the construction of models and simulation runs, and great interpretative flexibility of their outcome. Nevertheless, by relying on tacit expert knowledge and intuition, scientists were ultimately able to approach coherence and stability.