Abstract Understanding the behavior of 137Cs and 135Cs in soils and geological formations is of considerable interest in the context of nuclear accidents and nuclear waste repositories. Although the clay fraction is known to be responsible for sorption, there are still unanswered questions raised by the literature data concerning (i) the reversibility of the sorption process(es), (ii) the validity of the additivity rule (the overall distribution coefficient (Kd) for a radionuclide on a mixture of minerals is predicted from the distribution coefficients measured on individual minerals) and (iii) the validity of model transposition from dispersed systems to consolidated/intact systems. Because of these uncertainties, the validity of sorption models at equilibrium under in-situ conditions and for very long-term interaction is still pending. These different issues are studied in the present work for the Callovo–Oxfordian (COx) clay-rich rock Formation, which is under investigation in France as a geological barrier for a long-term nuclear waste repository. The work is based on sorption data measured on thirteen samples of different mineralogy taken from five different boreholes at several depths within the COx sedimentary layer. To our knowledge, it is the most extended Cs sorption dataset that has been published for a single clay formation in term of (i) sample locations (and thus natural variability), (ii) sorption conditions (powder dispersed in suspension, compacted powders and intact samples) and (iii) equilibration time (from one week to five years). Moreover, for the first time ever, radioactive Cs sorption results were compared to the natural distribution of non-radioactive Cs isotopes between pore water and the solid phase. The experimental system appeared to be in chemical equilibrium as much as can be expected for an ion-exchange reaction. More particularly, no kinetically-controlled process leading to partial Cs irreversibility was observed, in contrary to what was found in the literature for soils. This difference in behavior may be related to the difference in the illite studied, i.e. a soil-type illite which would be more altered than a sedimentary formation-type illite. No decrease in site capacity was observed between dispersed and intact/compacted states. A model based on exchange reactions with cations interacting with illite (frayed edge, type-II and planar sites) and mixed layer illite–smectite (I/S) (planar sites) using parameters published in the literature enabled the Kd variation to be described as a function of Cs concentration, the mineralogy of the samples, the change in water composition and the temperature (22–80°C). Our study clearly demonstrates that no frayed edge sites should be considered on the illite fraction of I/S, thus emphasizing the difference of sorption properties between an I/S mixed layer mineral and a corresponding mechanical mix of illite and smectite minerals. The robustness of the model was confirmed by data analysis describing the behavior of naturally-occurring Cs in the formation thereby demonstrating the effectiveness of the Cs sorption processes in a very long-time period prospective. Lastly, the model was used to predict the sorption of trace concentrations of Cs in the COx Formation on the time-scale relevant for nuclear waste disposal performance assessment. As expected, the retention was significant with Kd values ranging from 100 to 2000L/kg whatever the conditions that were probed and a simulation covering a period of over 105years could show that the COx Formation is an efficient barrier to prevent Cs transport from the storage facility to the surrounding environment.