Saline aquifers are choice targets for geological storage of CO2because of their storage potential andbecause these formations are not suitable for other uses. Geochemical modeling is an interesting tool toassess the geochemical behavior of CO2in the saline aquifer, including its dissolution in the brine andits interactions with minerals. Two key parameters which determine the confidence one can have in theresults of geochemical modeling are tested in this paper: (i) the establishment of the conceptual model,including the selection of the primary and secondary minerals expected to react; and (ii) the activitymodel and the associated thermodynamic databases to calculate the interaction energies within thesaline solution. In this study, we performed an analysis of a large set of CO2storage natural analogs, whichmakes it possible to identify the minerals that are likely to precipitate and dissolve during CO2-brine-rockinteractions. Interestingly, this analysis indicates a strong dependence of Dawsonite precipitation on theinitial sandstone mineralogy. Dawsonite can precipitate in lithic and feldspar rich sandstones but wasnot observed in quartz rich sandstones. These observations on mineral reactivity are used to establishreactivity conceptual models for three CO2storage case-studies in saline sandstone aquifers (Ketzin, InSalah and Snøhvit) and a methodology is proposed to evaluate the long-term geochemical reactivityof these saline aquifers as a result of CO2injection. Noticeable differences are obtained between thecase-studies as a function of the initial mineralogy and chemical conditions in the sandstones, whichhighlight that CO2mineral trapping can take place in a given storage site but can be almost absent inother storage sites. Regarding the activity model and the database, the Pitzer interaction model is rarelyused for simulating CO2geochemical behavior in saline aquifers despite the fact that more conventionallyused activity models are not valid for such salinities. A comparison between calculated mineral solubilityevolution with salinity versus experimental data is performed here using both B-dot and Pitzer activitymodels as well as six different databases. This comparison exercise shows that chemical interactionswithin saline solutions can only be reproduced using the Pitzer model, even though Pitzer databasesare still incomplete or are not coherent for a wide range of chemical species and temperatures. Thegeochemical simulations of CO2injection in Ketzin, In Salah and Snøhvit saline aquifers give divergentresults using different activity models and databases. A high uncertainty on the simulation results is thenlinked to the database choice and this study clearly stresses the need for a Pitzer database that can beconfidently used in all physical/chemical conditions found in deep sedimentary aquifers.