Elevated concentrations of atmospheric CO2 are implicated in global warming. Mitigation of this requires capture of CO, from fossil fuel power sources and storage in subsurface aquifers or depleted hydrocarbon fields. Demonstration projects and financial analysis suggest that this is technologically feasible. CO2 must retained below ground for 10(4) y into the future to enable the surface carbon cycle to reduce atmospheric CO2 levels. To provide robust predictions of the performance of disposal sites at the required timescale, one approach is to study natural CO2 accumulations, which give insight into rock-CO2-brine interactions over timescales of 10(3)-5.10(6) y. In contrast to geochemical modelling predictions, natural CO2 fields in the North Sea (Brae, Miller, Magnus, Sleipner), at 4.0 km and deeper, do not show the mineral products which are predicted to form. Calcite and feldspar still comprise 5-20% of the rock, and dawsonite is absent. SE Australian and Arizona reservoir sandstones also do not fit to geochemical predictions. A state of disequilibrium possibly exists, so that existing geochemical modelling is not capable of accurately predicting kinetic-controlled and surface-chemistry controlled mineral dissolution or precipitation in natural subsurface sandstones on the required timescales. Improved calibration of models is required. Geochemical evidence from laboratory experiments (months to years duration), or from enhanced oil recovery (30 y duration) are again too short in timescale. To help to bridge the 10(4) y gap, it may be useful to examine natural analogues (10(3)-10(6) y), which span the timescale required for durable disposal. The Colorado Plateau is a natural CO2 system, analogous to an hydrocarbon system, where 100 Gm(3) CO., fields occur, sourced from 0-5 Ma volcanics. Deep erosion has exposed the sediments which formed CO2 source, CO2 carrier, CO2 reservoir, CO2 trap, CO2 seal. Some very large CO2 traps are now exhumed, and some are currently leaking to form cool travertine springs at the surface. Natural examples at Salt Wash Green River, and at Moab Fault are briefly described. These show extensive bleaching of haematite which may be locally redeposited, carbonate cementation delta C-13-7 parts per thousand around point sources, and silica precipitation, which may seal leak-off on buried anticline crests. Accurate geochemical modelling of the long-term performance of CO2 storage sites, requires improved understanding of CO2 reaction paths and reaction rates with aquifer reservoirs and with overlying seals. Robust prediction of disposal site performance is not possible without this.