In-plant sheep carcass grading offer ram breeders opportunities to use carcass yield data from culls in genetic evaluations. This is a special case of two-stage selection where the second-stage measurement is not made on selection candidates, but on half-sib relatives that have been slaughtered.  Carcass yield data from 11,500 progeny of 257 sires were used to investigate the loss in eBV accuracy in this scenario.  All animals had pedigree recorded, weaning weight and weight of lean in the hindleg, loin and shoulder primal-cuts measured by VIAscan. Datasets were prepared where VIAscan measurements were retained for a random 5%, to 50% of animals, replicated five times.  eBVs were estimated for each dataset, along with datasets comprising 0% and 100% of VIAscan records retained.  Breeding values for all individuals without VIAscan records were regressed against their breeding value in the 100% dataset to determine the loss of accuracy from having records on a proportion of half-sibs rather than their own (plus their half-sibs) records.  Results indicate an asymptotic increase in accuracy of EBVs with increasing VIAScan records. When the proportion of animals with records reaches 50%, accuracy increases to 76.0%, 67.6% and 72.8% for VIAscan leg, loin and shoulder meat yield, respectively.