Bill Ruddiman (Climatic Change, 61, 261-293, 2003) recently suggested that early civilisations could have saved us from an ice age because land management over substantial areas caused an increase in atmospheric CO 2 concentration. Ruddiman suggests a decreasing "natural course" of the Holocene greenhouse gases concentrations and sea-level by referring to analogous situations in the past, namely the last three interglacials. An examination of marine isotopic stage 11 would perhaps make Ruddiman's argument even more thought-challenging. Yet, the hypothesis of a natural lowering of CO 2 during the Holocene contradicts recent numerical simulations of the Earth carbon cycle during this period. We think that the only way to resolve this conflict is to properly assimilate the palaeoclimate information in numerical climate models. As a general rule, models are insufficiently tested with respect to the wide range of climate situations that succeeded during the Pleistocene. In this comment, we present three definitions of palaeoclimate information assimilation with relevant examples. We also present original results with the Louvain-la-Neuve climate-ice sheet model suggesting that if, indeed, the Holocene atmospheric CO 2 increase is anthropogenic, a late Holocene glacial inception is plausible, but not certain, depending on the exact time evolution of the atmospheric CO 2 concentration during this period.The debate about the reason for the steady increase in CO 2 concentration between the Early Holocene (260 ppmv, 8000 years ago) and the pre-industrial era (280 ppmv, 300 years ago), recently spiced up by Bill Ruddiman's hypothesis that this increase could be due to early land management (Ruddiman, 2003), illustrates well two conceptions of understanding and forecasting climate dynamics.On the one hand, experts in biogeochemical cycles have shown that the increase in atmospheric CO 2 concentration during the Holocene is easily explained by a set of reasonable hypotheses about natural changes in land carbon sequestration, lysocline depth and sea-surface temperature in response to the orbital forcing. Seemingly, the only debate is about the respective contributions of these three factors (Indermühle et al., 1999;Brovkin et al., 2002;Joos et al., 2004). The MoBidiC model (Crucifix and Joos, 2004) simulates for its part no more than a 4 ppmv increase in CO 2 concentration in response to the orbital forcing during the Holocene (Figure 1, previously unpublished results). Unfortunately, these estimates are associated with considerable uncertainties. Thorne et al. (2004, submitted) recently termed 'structural uncertainties' errors associated with the hypotheses used in a model-based analysis process (it covers, in this case, aspects as various as inferring Climatic Change (2005) 69: 419-426