Measuring individual feed intake of fish in farms is complex and precludes selective breeding for feed conversion ratio (FCR). Here, we estimated the individual FCR of 588 sea bass using individual rearing under restricted feeding. These fish were also phenotyped for their weight loss at fasting and muscle fat content that were possibly linked to FCR. The 588 fish were derived from a full factorial mating between parental lines divergently selected for high (F+) or low (F–) weight loss at fasting. The pedigree was known back to the great grand-parents. A subset of 400 offspring and their ancestors were genotyped for 1,110 SNPs which allowed to calculate the genomic heritability of traits. Individual FCR and growth rate in aquarium were both heritable (genomic h2 = 0.47 and 0.76, respectively) and strongly genetically correlated (−0.98) meaning that, under restricted feeding, faster growing fish were more efficient. FCR and growth rate in aquariums were also significantly better for fish with both parents from F– (1.38), worse for fish with two parents F+ (1.51) and intermediate for cross breed fish (F+/F– or F–/F+ at 1.46). Muscle fat content was positively genetically correlated to growth rate in aquarium and during fasting. Thus, selecting for higher growth rate in aquarium, lower weight loss during fasting and fatter fish could improve FCR in aquarium. Improving these traits would also improve FCR of fish in normal group rearing conditions, as we showed experimentally that groups composed of fish with good individual FCR were significantly more efficient. The FCR of groups was also better when the fish composing the groups had, on average, lower estimated breeding values for growth rate during fasting (losing less weight). Thus, improving FCR in aquarium and weight loss during fasting is promising to improve FCR of fish in groups but a selection response experiment needs to be done. Finally, we showed that the reliability of estimated breeding values was higher (from+10% up to +125%) with a genomic-based BLUP model than with a traditional pedigree-based BLUP, showing that genomic data would enhance the accuracy of the prediction of EBV of selection candidates.