Several primary tetraploids with desirable attributes have been selected by Musa breeding programs. Diploid parental stocks have become available for further genetic improvement of the Musa genome. Hence, new genome manipulations should be tested before they become routine in breeding programs. Through comparison of the performance of triploid landraces, their primary tetraploid hybrids and secondary polyploid derivatives, plus diploid ancestors, it was found that the production of secondary triploids (TM3x) is more rewarding than developing secondary tetraploids. TM3x achieved significant high polyploid-parent heterosis for yield either by maximizing heterozygosity through crosses between unrelated parents, or by selection of linkats in hybrids derived from crosses between euploid full-sibs. There were significant differences in bunch weight between full-sib secondary polyploids, which suggested that specific combining ability should be re-defined considering not only a specific cross combination but also the individual performance within each cross. This paper proposes a crop-breeding strategy which takes into consideration the process of domestication of banana and plantain. Current data suggest that this type of evolutionary breeding approach may be feasible in Musa.