Agronomic evaluations demonstrated that a modification of the classical full-sib reciprocal recurrent selection (RRS-FS) which, in addition to crosses, uses S2 families evaluation (RRS-FS-S2) is more efficient than the classical method for developing high-yielding crosses between two varieties. The objective of this study was to investigate the changes in genetic diversity and structure after performing RRS-FS and RRS-FS-S2 selections. RRS-FS-S2 reduced more the variability, produced more differentiation between cycles of selection derived from the same materials but less between reciprocal populations, and produced a more clear change in the contribution of the parental lines than RRS-FS. On the other hand, the type of selection method did not have a considerable effect on the structure of the populations measured as departure of Hardy–Weinberg (HW) equilibrium at single markers and on linkage disequilibrium (LD) between pairs of markers. We identified some individual markers which were not in HW equilibrium in several populations probably due to genes favouring assortative mating. We also found pairs of markers which increased their LD with selection probably due to epistasis.