Banana (Musa acuminata) breeding programs aim to deliver new high-yielding cultivars, resistant to diseases and pests, and with high fruit quality. Their efficiency in terms of genetic gains per unit of cost and time, and their ability to release new cultivars rely on the development of many breeding applications based on quantitative genetics theory and on statistical analyses of numerous experimental data. In the present study, we took advantage of a banana breeding program of 2662 triploid hybrids obtained from a parental population of 48 cultivars reflecting Musa acuminata diversity, to document genotypic variation of several traits related to vegetative growth, morphological, phenological and yield characteristics. Triploid hybrids were obtained by crossing a diploid with a tetraploid genitor. The experimental set up was divided in 48 blocks nested within 12 plots, and hybrids were evaluated only once in the entire experiment. Since hybrids were not replicated across blocks or plots, the presence of 8 control plants per experimental unit allowed analysing all blocks in a single statistical model. To investigate the ability of the three statistical models to estimate accurately the genetic values of hybrids, a hundred traits were simulated. The model accounting for both block effects and diploid/tetraploid general and specific combining abilities was the most efficient to correct for experimental design effects. When applied to banana morphological and yield traits, the variance components demonstrated the predominance of the tetraploid parent contribution in the hybrid performance.