Global vegetation models are remarkably effective when considering large areas such as Europe. However, their accuracy at finer scales remains to be tested. In this paper, we validate the simulation of modern potential vegetation by the CARbon Assimilation In the Biosphere (CARAIB) model in Europe. Then, in order to evaluate the simulation of tree group distributions at a finer scale, in France, we present a comparison between observed distributions, distributions reconstructed from palynological data, and model simulated ranges. The results will help to validate past vegetation simulations. For this analysis, we use Bioclimatic Affinity Groups (BAGs), based on vegetation groups' climatic tolerances and requirements. The CARAIB model was adapted to simulate the net primary productivity (NPP), biomass and range of the arboreal BAGs. In Europe, at a 30′ latitude/longitude grid scale, simulated NPP of BAGs are used to define classes of vegetation as being present or absent, with a classification rule, based on Kappa statistics. In France, at a 10′ lat./long. scale, a second discriminant analysis, based on Classification And Regression Tree (CART), allows for a similar classification with BAG pollen percentages. At each palynological sampling site, we then compared the simulation to the reconstruction from pollen data. With 30′ lat./long. resolution, most thresholds that discriminate NPP into absence or presence classes are low, ranging from 1 to 77 g/m2. Agreement indices between observed and simulated distributions range from 0.4 to 0.83, with broad scale BAG potential patterns and boundaries being accurately simulated by CARAIB. In France, on the 10′ lat./long. scale, pollen percentages correctly account for BAG presence/absence despite non-linear pollen-vegetation relationships. Agreement ratios between observed and reconstructed patterns range from 0.53 to 0.95. At the 10′ lat./long. scale, the validation of simulated ranges with pollen data is reliable for 9 of 13 arboreal BAGs and acceptable for three more BAGs. The discrepancies highlight the gap between potential and actual distribution areas. The filling of simulated potential ranges, such as the Atlantic coast and near Mediterranean border, are uncompleted as actual ranges are limited by a number of climate and dispersal constraints related to competition as well as historical, geographical and anthropogenic factors. Our results suggest that the simulation of these constraints would be a major improvement for the CARAIB model.