Miscanthus has great potential as a biofuel crop because of its high productivity for low input. However, it is an undomesticated out-breeding perennial which makes the rapid development of commercial varieties demanded for carbon mitigation a challenging process. Therefore, the aim of this study was to access the potential of genomic selection (GS) to predict biomass yield and traits correlated to it, using a genome wide marker set. The GS predictive abilities (correlation of predicted and observed genetic values) across traits ranged from 0.76 for day of flag leaf production down to 0.06 for yield itself. To improve the yield prediction, a set of selection indices were built using between one and eight additional traits, aiming to minimise the phenotyping effort needed by breeding programs, while jointly maximising the response to selection and predictive ability through GS. The relative response to selection on the indices compared to yield alone were up to 16% higher, whereas GS on the selection indices resulted in predictive abilities that were up to six times higher than for yield. This is a substantial increase in the ability to make predictions about biomass yield and with further work could be used, for instance, to prioritise crosses, as well as to select seedlings directly from their marker profiles hence reducing breeding-cycle times.