This paper explores the relationship between past climate and prehistoric Mediterranean agriculture by adapting a process-based dynamic vegetation model to estimate potential agricultural productivity under climate scenarios that characterize the extremes of Mediterranean climate in the Holocene. We adapt LPJmL (the Lund-Potsdam-Jena-managed-land model [Bondeau et al., 2007]), a process-based dynamic vegetation model, to the modeling of potential agricultural productivity in the past. Calibrating this model for past crops and agricultural practices and using a downscaling approach to produce high spatiotemporal resolution paleoclimate data, we produce quantitative estimates of potential yields under past climatic conditions derived from four Holocene climatic extremes (warm/wet, warm/dry, cold/wet, and cold/dry) under two different assumptions (approximate high and low limits) about the intensity of agricultural practice. We here discuss this process with reference to a case study in Provence, examining the methodology and data requirements for modeling past agriculture using LPJmL and considering the implications of the range of variability in potential agricultural productivity under distinct climate conditions. We focus particularly on comparing the range of variability induced by climatic shifts with that achievable through changes in agricultural practices as a means of approaching questions of past vulnerability and resilience.< Réduire