Anthropogenic land use changes and hydroclimate fluctuations are generally described as the main control factors of soil erosion on centennial to millennial time scales, but their relative contribution on erosion trends is barely quantified yet. While past erosion dynamics can be inferred from lake sediment records, improvements in land surface models now allow the quantification of soil erosion on a variety of spatial scales ranging from plot to regional or even global scales. However, application of model approaches to long timescales is still at its dawn, limiting quantification of past soil erosion, investigation of scenarios, interpolation of data spatially and temporally, or testing hypothesis. Here we show how coupling paleo-environmental data and spatially distributed models of soil loss constrained by land cover data might help to assess past soil erosion dynamics and to quantify soil loss exports. The methodology has been tested on six alpine lake catchments that cover large altitudinal gradients (from 420 to 2494 asl) over the Holocene period. Our results show that soil erosion dynamics in the Alps seems to be characterized by two main regimes. 1) a long-term (i.e. pluri-millennial) slow increase in soil erosion that we attribute to progressive land use change, and 2) ‘short’ term (i.e. decennial to centennial) erosion crisis associated with faster land uses changes. Our results suggest that reconstructing past soil erosion requires accurate land cover change reconstructions. Our results further suggest that deforestation and land opening may have amplified the effect of precipitations on soil erosion. Coupling model and paleolimnological approaches should hence open new avenues to assess in a more integrative way mass fluxes and stocks within lake catchments systems over long-term periods.