Résumé
Using the pollen sequence of La Grande Pile XX (France), we review problems with the application of transfer functions in paleoclimatic reconstructions. One of them is to find modem analogues for the herbaceous vegetation of the cold periods. We propose a method to distinguish between steppes and tundra vegetations for which the modems are only partial analogues of the glacial periods. Another method to solve these problems is based on constraining by insect remains. The two methods provide coherent reconstructions. The results are also compared with other paleodata. There is a good correlation with the six cold Heinrich events between 70 and 15 ka B.P. A cooling event during the Eemian period (marked by high percentages of Taxus) at about 125 ka B.P. needs still to be correlated with high resolution ocean and ice cores.