Detection of loci involved in variation of agronomic traits is an important issue leading to marker assisted selection. In this field, linkage analyses based on pedigree data limit the study to a low number of alleles and a limited number of recombinations (meioses). That is why QTL (Quantitative Trait Locus) analyses encounter difficulties to estimate precisely gene location. In consequence, association studies based on more complex populations such as core collections, natural or selection populations are applied to refine gene location estimates. The challenge is then to detect functional associations while not taking into consideration false positives due to demographic history. This strategy has been applied to Sorghum (Sorghum bicolor ssp bicolor), a preferentially autogamous and annual crop. A first genome scan with high throughput markers (DArT) has been performed on a core collection of 190 samples representative of the worldwide diversity of cultivated sorghum. According to 935 markers, the collection is structured both by race and origin,which is congruent with genetic analyses done with RFLP and SSR. Among those DArT markers, 330 non redundant were mapped, covering the whole genome with approximately one marker each 3cM. No association between markers was found beyond this window of 3 cM and rarely between markers mapped on different chromosomes. These results suggest that statistical associations are due to physical linkage only, as expected to undertake some association studies. (Texte intégral)