Description: We compare Holocene tree-cover changes in Europe derived from a transient MPI-ESM1.2 simulation with high spatial resolution LPJ-GUESS time-slice simulations and pollen-based quantitative reconstructions of tree cover based on the REVEALS model. The dynamic vegetation models and REVEALS agree with respect to the general temporal trends in tree cover for most parts of Europe, with a large tree cover during the mid-Holocene and a substantially smaller tree cover closer to the present time. However, the decrease in tree cover in REVEALS starts much earlier than in the models indicating much earlier anthropogenic deforestation than the prescribed land-use in the models. While LPJ-GUESS generally overestimates tree cover compared to the reconstructions, MPI-ESM indicates lower percentages of tree cover than REVEALS, particularly in Central Europe and the British Isles. A comparison of the simulated climate with chironomid-based climate reconstructions reveals that model-data mismatches in tree cover are in most cases not driven by biases in the climate. Instead, sensitivity experiments indicate that the model results strongly depend on the tuning of the models regarding natural disturbance regimes (e.g. fire and wind throw). The frequency and strength of disturbances are – like most of the parameters in the vegetation models – static and calibrated to modern conditions. However, these parameter values may not be valid during climate and vegetation states totally different from todays. In particular, the mid-Holocene natural forests were probably more stable and less sensitive to disturbances than present day forests that are heavily altered by human interventions. Our analysis highlights the fact that such model settings are inappropriate for palaeo-simulations and complicate model-data comparisons with additional challenges. Moreover, our study suggests that land-use is the main driver of forest decline in Europe during the mid- and late-Holocene.