There has been a gradual evolution in the way that palaeoclimate modelling and palaeoenvironmental data are used together to understand how the Earth System works, from an initial and largely descriptive phase through explicit hypothesis testing to diagnosis of underlying mechanisms. Analyses of past climate states are now regarded as integral to the evaluation of climate models, and have become part of the toolkit used to assess the likely realism of future projections. Palaeoclimate assessment has demonstrated that changes in large-scale features of climate that are governed by the energy and water balance show consistent responses to changes in forcing in different climate states, and these consistent responses are reproduced by climate models. However, state-of-the-art models are still largely unable to reproduce observed changes in climate at a regional scale reliably. While palaeoclimate analyses of state-of-the-art climate models suggest an urgent need for model improvement, much work is also needed on extending and improving palaeoclimate reconstructions and quantifying and reducing both numerical and interpretative uncertainties.