Résumé: Fillet yield, the proportion of edible fillet relative to body weight, is a major trait to improve in fish sold processed, as it has a direct impact on profitability and can simultaneously decrease the environmental impact of producing a given amount of fillet. However, it is difficult to improve by selective breeding, because it cannot be measured on live breeding candidates, its phenotypic variation is low, and, as a ratio, it is not normally distributed and a same change in fillet yield can be the result of different changes in fillet weight and body weight. Residual headless gutted carcass weight (rHGCW) is heritable and highly genetically correlated to Fillet% in rainbow trout, and can be predicted by the ratio of abdominal wall thickness to depth of the peritoneal cavity (E8/E23), measured on live fish by ultrasound tomography. We selected broodstock based on rHGCW, measured on sibs of the selection candidates, on ultrasound measurements (E8/E23) measured on the selection candidates, or a combination of both. Seven broodstock groups were selected: fish with 15% highest (rHGCW+) or lowest (rHGCW-) EBV for rHGCW, with 15% highest (E8/E23+) or lowest (E8/E23-) EBV for E8/E23, with both rHGCW+ and E8/E23+ (Both+) or rHGCW- and E8/E23- (Both-), or with close to zero EBVs for both traits (Mid). Seven corresponding groups of offspring were produced and reared communally. At harvest size (1.5 kg mean weight), 1,561 trout were slaughtered, measured for the traits of interest, and pedigreed with DNA fingerprinting. Offspring from groups Both+, rHGCW+ and E8/E23+ had a higher EBV for rHGCW than the control group, while down-selected groups had a lower EBV. Looking at the phenotypic mean for Fillet% (correlated response), up-selected fish had more fillet than down-selected fish. The highest difference was between Both+ (69.36%) and Both- (68.20%), a 1.16% units difference in fillet percentage. The change in Fillet% was explained by an opposite change in Viscera%, while Head% remained stable. Selection using sib information on rHGCW was on average more efficient than selection using the candidates' own E8/E23 phenotypes, and downward selection (decreasing Fillet%) was more efficient than upward selection.