Today, a monitoring network measures atmospheric CO2 at high temporal and spatial resolution. Atmospheric transport models then calculate regional CO2 source and sink fluxes. Prior to this instrumental period, archived air, firn air, and air trapped in polar ice are the only direct atmospheric archives to reconstruct past CO2 changes. Only ice from Antarctica allows reliable CO2 measurements, either from classical ice cores or outcropping ice, while Greenland records are subject to in situ production. They provide high-resolution and high-precision CO2 reconstructions up to 800,000 years back in time. Ice core records have revealed an intimate connection between CO2 variations and major changes in Earth’s climate and have fundamentally shaped the community’s view of the global carbon cycle. Knowing the concentration of past atmospheric CO2 and the other greenhouse gases is key to provide the radiative forcing for climate simulations. Ice core reconstructions broadly fulfilled this task. On the contrary, we are far from a coherent understanding of the mechanisms driving these changes. Analyzing phase relations between CO2, other ice-core derived species, and proxies from marine sediment cores allow for the identification of factors likely responsible for the observed CO2 changes. Specifically, the strength of the Atlantic overturning circulation and Southern Ocean upwelling are thought to be key players. However, the observed CO2 changes cannot uniquely be related to a specific process. Here, stable carbon isotope analysis on CO2 extracted from ice provides additional constraints as any process leads to isotope fractionation of the reservoir. Analytical progress during the last decade affords us with a growing data set on this long-awaited parameter. This presentation provides a state-of-the-art overview on ice-based CO2 and its carbon isotopic signature focusing both on the long-term orbital changes as well as rapid changes documented during the last deglaciation.