Long-term perspectives on climate- and human-induced shifts in plant communities and tree line in mountains are often inferred from fossil pollen records. However, various factors, such as complex patterns of orographic wind fields and abundant insect-pollinated plants in higher altitudes, make pollen-based reconstruction in mountain regions difficult.Over the last decade the Landscape Reconstruction Algorithm (LRA) - a model-based approach in reconstruction of vegetation – has been successfully applied in various parts of the globe. However, evaluation of its effectiveness in mountain ranges is still limited. The present study assesses the extent to which the LRA approach helps quantify the local changes in vegetation cover at Vicdessos valley in northern French Pyrenees as a case study. In the study area well-dated sediment cores are available from eight bogs and ponds, 6–113 m in radius, located above the current tree line. We first use a simple simulation experiment to evaluate the way how pollen records from “landscape islands” (mountain tops and plateaus) would represent local vegetation and to clarify important factors affecting the LRA-based reconstruction in a mountainous region. This study then uses pollen records from these sites and vegetation and land-cover data both within a 50-km radius around the Vicdessos valley and within a 2-km radius from each site for evaluation of the REVEALS- and LOVE-based reconstruction of the regional and local plant cover, respectively, in the LRA approach. The land-cover data are complied for coniferous trees, broadleaved trees and non-forested areas from the CORINE and historical maps in three time windows: 1960–1970, 1990–2000 and 2000–2013.Major findings are as follows. (1) Accuracy of the regional vegetation estimates affects the reliability of the LRA-based reconstruction of vegetation within a 2-km radius; use of the CORINE data as input to the LOVE model improves reliability of the results over the use of the REVEALS-based estimates of regional vegetation. This implies that a systematic selection of pollen data only from sites above the tree line is problematic for estimating regional vegetation, and thus the entire LRA process. (2) Selection of the dispersal models for pollen transport (i.e. the Langrangian Stochastic Model vs. Gaussian Plume Model) does not affect significantly the LRA-based estimates at both the regional and local scales in the study area. (3) The LRA approach improves the pollen-based reconstruction of local vegetation compared to pollen percentage alone in northern Pyrenees. Although further empirical and simulation studies are necessary, our results emphasize the importance of site selection for the LRA-based reconstruction of vegetation in mountain regions.