Studies addressing the variation of mating system between plant populations rarely account for the variability of these parameters between individuals within populations, although this variability is often non-negligible. Here, we propose a new direct method based on paternity analyses (Mixed Effect Mating Model) to estimate individual migration (m(i)) and selfing rates (s(i)) together with the pollen dispersal kernel. Using this method and the KINDIST approach, we investigated the variation of mating system parameters within and between three populations of Fagus sylvatica along an elevational gradient. Among the mother trees, s(i) varied from 0% to 48%, m(i) varied from 12% to 86% and the effective number of pollen donors (Nep(i)) varied from 2 to 364. The mating patterns differed along the gradient, the top population showing higher m and lower s, and a trend to higher Nep than the bottom populations. The phenological lag shaped long-distance pollen flow both within population (by increasing m(i) at mother-tree level) and between populations (by increasing m at high elevation). Rather than the mate density, the canopy density was detected as a major mating system determinant within population; it acted as a barrier to pollen flow, decreasing the proportion of long-distance pollen flow and increasing s(i). Overall, the effects of ecological factors on mating system were not the same within vs. between populations across the gradient, and these factors also differed from those traditionally found to shape variation at range-wide scale, highlighting the interest of multiscale approaches.