Levels of genetic diversity were estimated for 8 forest trees of French Guiana which were selected because of their contrasting life history traits. Multilocus methods are used as RAPDs or AFLPs in order to assess genetic diversity at the whole genome level. However these markers suffer from a major drawback: they behave generally as dominant markers. Theoretical calculations were made to compare two different methods to assess genetic diversity with dominant markers, the phenotypic diversity (Hp) that considers presence and absence of fragments versus the genotypic diversity (Hg) that considers the alleles responsible for the presence or the absence of the bands. It is shown that these values (Hp and Hg) can only be compared in a limited number of cases, according to the frequency of the bands. The genetic monitoring of gene diversity is based on a two step sampling procedure: (i) the intra-locus sampling, e.g. sampling of trees and populations and (ii) the inter-locus sampling, e.g. sampling of loci within the genome. The associated components of sampling variance were estimated using resampling methods in order to make species comparisons. It is shown for both markers that the inter-locus variance exceeds by far (5 to 10 times) the intra-locus variance. Furthermore both components of the sampling variance are much higher for RAPDs than for AFLPs, even if both markers provide congruent rankings of the species. There are striking differences in the level of phenotypic and genotypic diversity among the species. (Texte intégral)