For elephant grass breeding for bioenergy purposes, selection should be based on repeated harvests to maximize the selection efficiency. Selecting genotypes with high total dry biomass, associated with high adaptability and stability, is extremely important. Thus, the objective of this study was to estimate the minimum number of harvests required for total dry biomass to enable an accurate selection of elephant grass clones in multiple environmental trials. Additionally, the relative performance of genetic values (HMRPGV) and centroid methods were used to cluster and rank the best clones on the basis of their total dry biomass, adaptability, and stability. Nineteen clones and two checks were evaluated in six environments. Two or four harvests were performed in each environment. A mixed model methodology (REML/BLUP) was used to estimate the variance components and predict the genotypic values. The repeatability coefficient for total dry biomass was 0.20. The accuracies of performing two and four harvests were 59% and 72%, respectively. The increase in selection efficiency by performing two, four, and seven harvests instead of just one was 29%, 57%, and 76%, respectively. Finally, the clones were clustered into four classes according to their total dry biomass, adaptability, and stability. We conclude that at least seven harvests are required to achieve an accuracy of 80%, and that the HMRPGV and centroid methods are practical, visual, and reliable for simultaneously clustering, ranking, and selection of clones based on their adaptability, stability, and total dry biomass production.