These relationships probably vary as a function of the population structure. They are generally strong if there is a serious gametic disequilibrium. To proceed from this reflection, however, we must tackle specific questions: Are the different types of molecular markers equivalent? Are strong structures at the molecular scale, which control the gametic disequilibrium generalized on the genome as a whole, systematically associated with strong structures for agronomic characters? Do the structures at these two levels therefore coincide?Several methods have been proposed to construct the core collections. As a general rule, they are based on a stratification of the base collection and then on a random sample within each group thus defined, according to various modalities. When data are available for morphoagronomic evaluation, it is useful to consider the structure of the genetic diversity established with the help of genetic markers in order to maximize the agronomic diversity and minimize the loss of alleles that are rare overall but occasionally significant.The constitution of a core collection must therefore be founded on an excellent description of the populations and on a sound understanding of their structure. Statistical tools are indispensable here. For each type of marker, it is important to use a measure of difference between the relevant taxonomic units (individuals or populations) from the perspective of their mathematical properties and their interpretation in genetic terms. In the same way, the choice of algorithms of representation of dissimilarities must be based on an equilibrium between maximum efficiency and minimum complexity, irl order to deal with large tables covering more than 100 individuals. By comparing the structures observed with the help of various types of descriptors, we can envisage overall the organization of a plant's genetic diversity. This question can be tackled in various ways, among others, by looking for structures common to two or several trees. In these conditions, how can we construct consensus trees or common minimum trees? What is their reliability and their biological significance?This work provides the elements of an answer to these questions, beginning with the study of eleven types of plants, chosen .so as to cover a wide range of biological characteristics (perennial, annual, autogamous, allogamous, etc.): asian rice, banana, cacao, cassava, citrus, coconut, coffee, pearl millet, rubber tree, sorghum, and sugarcane. Three methodological chapters complement these studies. The first is devoted to the use of biological and molecular markers to analyse the diversity of collections, the second addresses data analysis, and the third describes a method for constituting core collections based on maximization of variability.