By trapping infrared radiation, atmospheric CO2 contributes significantly to the greenhouse warming of the planetary surface. Hence, it is thought to have played a key role in the evolution of the Earth's climate over geological time. The history of atmospheric CO2 is available only for the last few hundred thousand years from the analysis of the air trapped in cores of ice. Therefore, data regarding Pre-Pleistocene atmospheric CO2 must be derived from proxies. These provide indirect estimates of atmospheric CO2 and are much less reliable than ice-core data. The main proxies used to reconstruct atmospheric CO2 are: the 13C isotopic fractionation of marine organisms, the paleo-pH recorded in the boron isotopic composition of ancient carbonates, the stomatal density of fossil leaves and the 13C isotopic composition of paleosols. For Paleozoic times paleosols have been the main source of data but these are generally rather imprecise. Consequently, for this period geochemical models are useful to make first order estimates of atmospheric CO2 levels, as well as to help explain its temporal variation. Such models describe the geochemical cycles of several elements - the core being the carbon cycle - by writing budget equations for these elements in the framework of box models. They are often constrained by isotopic data. In the following we first summarize the basic principles of these models and then illustrate two applications: (1) changes in Paleozoic atmospheric CO2 and (2) changes in the carbon cycle across the Permo-Triassic boundary.