Genome size variation occurs within and across generations in Musa spp., which reduces the predictive accuracy of parental performance on progeny value for yield and other traits with complex inheritance. Parental selection through progeny testing of prospective parents is required to achieve further genetic gains. This was carried out in this study, using a factorial mating design involving five 4x females and five 2x males. Genetic differences among offspring families were essentially due to differences in additive effects of the parents. Thus, little recombinative heterosis can be expected upon 4x-2x cross-breeding, and breeding strategies should target the development of 4x and 2x cultivars by accumulation of favorable alleles through recurrent selection within each ploidy pool. Offspring yield was positively correlated with parental GCAs but not with mid-parent values. Hybrid performance was also associated but not significantly correlated with genetic similarity indices based on both pedigree and molecular data. This study further suggests that current genetic models may not be adequate for populations with intergeneration genome size polymorphism.