Most banana cultivars are triploid, sterile and seedless but there are also wild diploid banana highly fertile and parthenocarpic diploids more or less fertile. Banana production is severely threatened by pests and diseases. Wind damage also affects yield. Two approaches were coupled to manipulate ploidy for producing improved varieties. The evolutionary approach aiming to improve the existing triploid cultivars is based on triploid x diploid crosses generating diploids and tetraploids; these tetraploids were then crossed with other diploids to create secondary triploids. The 'reconstructing approach' aims to create new triploids from putative ancestral diploids; it couples in vitro colchicine induced tetraploidy from selected diploids; the resulting tetraploids are then crossed with other diploids to synthesize sterile and parthenocarpic triploids. Several hundreds of hybrids have been generated through embryo rescue. Flow cytometry is routinely used to determine their ploidy. The triploid x diploid scheme produces mainly diploid and tetraploid hybrids indicating that triploid banana can generate haploid and non- reduced triploid gametes. However, the progenies of tetraploid x diploid crosses are triploid. The most interesting achievements include segregating progenies for dwarfism obtained from a dwarf plantain; this result indicates that dwarfism in this cultivar has a genetic basis although dwarf off types are frequently generated from tissue-cultured commercial Cavendish cultivars. The agronomic performances of dwarf and normal triploid hybrids will be presented. On the basis of these first results on plantain banana, on-going research combining conventional and mutation breeding to develop various dwarf hybrids (dessert and cooking types) will be discussed. (Texte intégral)