Genetic diversity and structure of 12 populations of Eryngium alpinum L. were investigated using 63 dominant amplified fragment length polymorphism (AFLP) and seven codominant microsatellite (48 alleles) markers. Within-population diversity estimates obtained with both markers were not correlated, but the microsatellite-based fixation index Fis was correlated with both AFLP diversity indices (number of polymorphic bands and Nei's expected heterozygosity). Only AFLP diversity indices increased with the size of populations, although they did not significantly differ among them (Kruskall-Wallis test). The discrepancy between AFLPs and microsatellites may be explained by a better coverage of the genome with numerous AFLPs, the higher mutation rates of microsatellites or the absence of significant difference among within-population diversity estimates. Genetic differentiation was higher with AFLPs (theta=0.40) than with microsatellites (theta=0.23), probably due to the higher polymorphism of microsatellites. Thus, we considered global qualitative patterns rather than absolute estimates to compare the performance of both types of markers. On a large geographic scale, the Mantel test and multivariate analysis showed that genetic patterns were more congruent with the spatial arrangement of populations when inferred from microsatellites than from AFLPs, suggesting higher homoplasy of AFLP markers. On a small spatial scale, AFLPs managed to discriminate individuals from neighboring populations whereas microsatellites did not (multivariate analysis), and the percentage of individuals correctly assigned to their population of origin was higher with AFLPs than with microsatellites. However, dominant AFLPs cannot be used to study heterozygosity-related topics. Thus, distinct molecular markers should be used depending on the biological question and the geographical scale investigated.