Vegetation–climate relationships are often different at varying spatial scales, which is seldom taken into account in reconstructing past climate changes from fossil pollen spectra represent more local vegetation composition but using regional or extra-regional modern pollen–climate calibration set. In this paper, we employ 2620 surface pollen spectra and six pollen sequences from continental East Asia to reconstruct Holocene climate with modern analogue technique. A novel data set of vegetation sensitivity index (VSI) is introduced to examine vegetation–climate relationships in 0.5° search windows around the fossil sites (local) and in the tailored calibration sets (regional). Then, we compare them to the explanatory abilities of (reconstructed) climate variables on explaining the pollen variance in calibration sets and in the fossil pollen data sets using constrained ordination, cross-validation and significance test. By this procedure that we called local–regional–fossil comparison, we can better determine which variable reconstruction is valid and useful. In our cases, moisture variable reconstructions in temperate forest-steppe ecotone at the East Asian summer monsoon margin are reliable. Consistent Holocene moisture variations with a strong monsoon influence during 8.6–4.0 cal ka BP which can help us to better interpret human adaptations along the Great Wall. However, quantitative climate reconstructions at some sites are less reliable due to highly heterogeneous vegetation sensitivities surrounding the site, particularly for areas with complicated topographical contexts. We stress that vegetation sensitivities at local and regional scales both need to be investigated before conducting quantitative climate reconstruction.