The concept of CO2 storage relies on the long-term sealing properties of both the geological trap and the wells needed to inject and monitor CO2. Well integrity, a classical topic in the oil and gas industry, is thus critical for the performance of any CO2 storage complex in terms of containment. Thanks to the very low permeability of cement (typically less than 0.1 mDarcy); a properly cemented well ensures hydraulic isolation between reservoirs layers and shallow aquifers. Moreover, such low matrix permeability limits the cement/ CO2 interactions over the active period of a storage complex (of the order of 100 years) to a few meters. Leaks from a cased and cemented well, if any, are known to occur only through defects: mud-channel (in case of poor cement placement), cracks within cement and more importantly micro-annulus at the casing/cement or/and cement/formation interfaces. This last category of defects can lead to substantial leakage rate. Its importance has been recognized by the oil and gas industry since the 1960's leading to the study of cement "bonding" properties. In the scope of CO2 storage, the understanding, modeling and monitoring of the occurrence of micro-annulus becomes of prime importance. We analyze the complete loading history of a cemented completion from cement placement to routine well operations. Further to classical failure type assessment used in the oil and gas industry (i.e. fail/no fail, good cement/bad cement), we aim at quantifying the vertical extent, azimuthal coverage and width of the created defects to adequately transform failure types into leakage pathways. Such a prediction of connected defects/leakage pathways along a cemented well imposes to consistently integrate the effects of lithology, geomechanics, cement placement (fluid loss, hydration), completion design and knowledge of pressure and thermal variation during the life of the well. The modeling of such a problem can be made tractable by recognizing the intrinsic hierarchy of lengthscales of a cemented well (i.e. the cement annulus is much thinner than the well dimension). The original three-dimensional problem is reduced to a much simpler two-dimensional one, which in turn can even be further reduced to a one-dimensional configuration in a lot of practical cases. Typical cases of interface debonding due to well de-pressurization and thermal cooling taking place after cement placement are carefully analyzed. Furthermore, we specially focus on injectors. Despite the use of all current best practices during well construction, the injection in itself can lead to the propagation of a debonding crack between cement and casing or cement and formation due to the high pressure generated at the perforations level. Such a problem has already been reported in hydraulic fracturing operations, and is a reasonable explanation of observed well leaks for injectors. A consistent model predicting the initiation and propagation of interface debonding during injection operations is then compared to carefully designed laboratory experiments. Such experiments also confirm that the azimuthal coverage of the interface debonding is only partial (i.e. less than 360°), an observation consistent with cement evaluation logs acquired on CO2 injectors. Finally, best practices to achieve and retain well integrity of CO2 injectors are highlighted from a careful examination of the results of both the model and the experiment. © 2011 Published by Elsevier Ltd.