Cultivated bananas are derived from hybridization events among species and subspecies of the genus Musa that differentiated in various regions and archipelagos of tropical South-East Asia. These hybridizations generated diploid and triploid inter(sub)specific hybrids with impaired fertility. Some of them, having seedless parthenocarpic fruits, were selected by humans and further dispersed through vegetative propagation. The genomes of these domesticated bananas are expected to have gone through few/very limited rounds of recombination and to be organized in a mosaic of large blocks of sequences from different ancestral origins. To characterize mosaic genome structures of diploid bananas, we generated a set of SNPs from RNAseq data of 24 selected seedy (wild) and parthenocarpic (cultivated) diploid banana accessions. We applied multivariate (COA) and SNP clustering approaches to assign alleles to ancestral banana groups. Group assigned alleles were then used to locally infer the ancestral origin of genomic regions along the 11 banana chromosomes. The results showed instance of relatively simple hybrid mosaic structures for some accessions derived from two ancestral groups that corresponded to two Musa acuminata subspecies. Other cultivated accessions showed more complex mosaics and involved more than two ancestors. Interestingly, the results also suggested that at least one unknown ancestral group, in addition to the four main groups already identified; contributed to the studied cultivated bananas. Globally, this study showed that some of the domesticated banana cultivars originated from a more complex history of hybridization events than previously thought. Deciphering the mosaic structure of banana genomes will improve our understanding of banana domestication and help breeding programs in their strategies to reconstruct improved hybrids.