During underground carbon dioxide (CO2) storage operations in deep reservoirs, the CO2 can be trapped in three ways; – as “free” CO2, most likely as a supercritical phase (physical trapping); – dissolved in formation water (hydrodynamic trapping); – precipitated in carbonate phases such as calcite (mineral trapping). This study focuses on the reactions between CO2, porewater and host rock. The aim of this work was to provide a well-constrained long-term laboratory experiment reacting known quantities of minerals with CO2-rich fluids, in order to try and represent situations where CO2 is being injected into lithologies deep underground. The experimental results can then be used as a test case with which to help validate predictive geochemical computer models. These will help improve our ability to predict the long-term fate of carbon dioxide (CO2) stored underground. The experiment, though complex in terms of equipment, ran for approximately 7.5 months. The reacted material was then examined for mineralogical changes and the collected fluids analysed to provide data on the fate of the dissolved species. Changes were readily observable on the carbonates present in thestarting material, which matches well with the observed trends in the fluid chemistry. However, although changes in silica concentrations were seen in the fluid chemistry no evidence for pitting or etching was noted in the silica bearing phases. Modelling of the experimental systems was performed using the BGS coupled code, PRECIP. As a general conclusion, the model predictions tend to over estimate the degree of reaction compared with the results from the experiment. In particular, some mineral phases (e.g. dawsonite) that are predicted to form in large quantities by the model are not seen at all in the experimental system. The differences between the model predictions and the experimental observations highlight the need for thermodynamic and kinetic data to be available under appropriate conditions (pH, and chemical composition of the fluid as well as temperature, and pressure), as extrapolation or “best guesses” may lead to errors being induced in the predictions. These errors and gaps in the data become obvious when comparing model predictions with experiments which serves to emphasise the importance of having “test cases” with which the models can be validated.