We investigated the relationships between the distribution of seven plant communities, using stepwise multiple logistic regression at a " Natura 2000 " site in the southern French Alps near Gap (Hautes-Alpes Department). The models are driven by topographical and bioclimatic parameters, geological substrate, and grazing pressure. These direct and indirect environmental gradients were derived from a Digital Elevation Model (DEM), with a resolution of 50 m. The accuracy of predictive models was tested using resampling techniques and ROC-plot diagnostics. Vegetation patterns in the study area were associated most strongly with bioclimatic variables, but also significantly with geological substrate and grazing pressure. The predicted distribution of communities, which covered 80% of total area, was overlaid on a map of independent observed natural habitat distribution data. The accuracy was assessed using the Kappa-statistic (?). The overall degree of coincidence between simulated and observed patterns was good (? = 0.62). Information on land-use history was identified as a major factor that could significantly improve the modelling of communities. This approach can provide an aid to decision-making and monitoring of Natura 2000 sites.