I discuss the role of model/data comparisons for past climate changes and use of such comparisons for enhancing credibility in future projections. I outline a framework in which data synthesis combined with suitable modelling targets should be able to reduce uncertainty in both. By focusing on areas that the latest Intergovernmental Panel on Climate Change assessment report (IPCC AR4) highlighted as being particularly uncertain in future projections, or where current models produce a very wide range of responses, the relevance of palaeoclimate data could be greatly enhanced. Specific targets include: the long-term behaviour of El Niño events and the potential response to volcanic and solar forcing; the variability of subtropical rainfall and the extent of the Hadley Circulation and their response to orbital and high-latitude forcing; ice sheet responses on sub-millennial timescales; multidecadal changes in the North Atlantic ocean circulation and, certainly, overall climate sensitivity. In each case, I highlight data synthesis steps and modelling approaches necessary for reducing the uncertainty. In particular, I stress the need for coordinated model simulation archives that are conformal to those used in simulations of the 20th century and beyond and the consistency of models used for past and future climate simulations. Published in 2009 by John Wiley & Sons, Ltd.