Carbon dioxide (CO2) geological storage relies on safe, long-term injection of large quantities of CO2 in underground porous rocks. Wells, whether they are the conduit of the pumped fluid or are exposed to CO2 in the storage reservoir (observation and old wells) are man-made disturbances to the geological storage complex, and are thus viewed by some as a possible risk factor to the containment of the injected CO 2. Wells are composite structures, with an inner steel pipe separated from the borehole rock wall by a thin cement sheath (∼2 cm) that prevents vertical fluid migration. Both carbon steel and cement react in the presence of CO2, although evidence from production of CO2-rich fluids in the oil and gas industry and from lab experiments suggests that competent, defect-free cement offers an effective barrier to CO2 migration and leaks. However, reactivity of cement and steel may result in CO2 migration pathways degrading over time, thus in the leakage risk increasing during the life of the storage project. The issue then becomes how to best integrate preventive verification of zonal isolation/well integrity in the storage site monitoring plan. An analysis of the order of magnitude of possible CO2 leaks, and of their path to potable aquifers or the atmosphere, is also necessary to optimize the assurance (mitigation) monitoring of the storage site. Evidence gathered during the MovECBM project indicates that migration of small quantities of CO2 happened during injection in a coal seam in Southwest Poland. The evidence, gathered from casing and cement logging as well as soil gas monitoring over a 3-year period, was coupled with laboratory testing and extensive modeling of the chemo-mechanical behavior of cement and steel to determine if CO2 migration might have been responsible of the observed behavior. The three lines of evidence were: the detection of very small CO2 fluxes, coupled with less controversial helium concentration in soil; the occurrence of a thin pathway at the interface between cement and casing; and the change in mechanical properties of cement, suggestive of partial carbonation. Whereas the observations suggest that limited CO2 migration might have happened in the well, they are by no means proof that the migration did happen. Nonetheless, the integration of measurement and modeling yields important lessons for wellbore monitoring. First, it puts a probable ceiling on the order of magnitude of expected leaks from reasonably well-cemented wells at around 100 metric tons per year (less than 0.05% of the injected mass in a well like Sleipner or In Salah). It also suggests that cement may be a very effective leak detector: exposure to CO2 modifies its mechanical properties, which in turn can be detected using cement evaluation logs. Finally, coupling with dispersion modeling suggests the precision and accuracy required from soil gas and atmospheric monitoring, as well as the placement of sampling points; it also suggest that hysteresis, due to the accumulation in CO2 in surface aquifers and to the time required for it to be transported to the survey points, may delay initial detection; the same hysteresis may at the same time prolong the occurrence of CO2 shows long after the leak has stopped. © 2010 Elsevier Ltd. © 2011 Published by Elsevier Ltd.