Aim: This paper presents a probabilistic method for the characterization of pollen taxa using attributes, and for the reconstitution of past biomes. The probabilities are calculated on the basis of European floristic and pollen databases sufficiently large and exhaustive to provide robust estimates. Location: The analysis is based on data from approximately 1000 sites throughout Europe. Method: We use all the pollen data from the European Pollen Database (EPD), which contains about 50 000 pollen assemblages distributed across Europe and covering the period from the Last Glacial Maximum to the present. Using existing floras, each pollen taxon has been characterized by allocating one or more modes of several attributes, chosen according to the biogeography and phenology of the taxon. With this information, conditional probabilities are defined, representing the chance of a given attribute mode occurring in a given pollen spectrum, when the taxa assemblage is known. The concept of co-occurrence is used to provide a greater amount of information to compensate for difficulties in the identification of pollen grains, allowing a better interpretation when there is little diversity in the pollen assemblage. Results: The method has been validated using a dataset of modern samples against existing methods of biome classification and remote sensing data. An application is proposed in which the new method is used to produce biomes for pollen data 6000 years ago. This confirms previous results showing an extension of the deciduous forest to the north, east and south, explained by milder winters in western and northern Europe, and cooler and wetter climate in the Mediterranean region. Conclusion: The results show the new method to be efficient, reliable and flexible and to be an improvement over the previous method of biomization. They will be used to test simulations of earth system models running on periods with climate significantly different from the present day, enabling a robust test of the validity of applying these models to the future.