Genomic predictions of transmitting ability (GPTAs) for residual feed intake (RFI) were computed using data from 4,621 42-d and 202 28-d feed intake trials of 3,947 US Holsteins born 1999–2013 in 9 research herds. The 28-d records had 8.5% larger error variance than 42-d records and received less weight (0.92 vs. 1.0) in the evaluation. The RFI averages were already adjusted to remove phenotypic correlations with milk energy output, metabolic body weight, and body weight change and for several environmental effects including other nutrition experiments during the feed intake trials. Traditional breeding values (BVs) for RFI of 74.3 million Holsteins were obtained by an animal model that also included effects for age-parity group, trial date, herd management group, permanent environment, herd-sire interaction, and regressions on inbreeding and on genomic evaluations for milk energy and body weight composite (BWC). The milk energy and BWC terms were specified with the intent to remove positive genetic correlations that remained after the phenotypic correlations were removed. Estimated heritability was 0.17 and repeatability across lactations was 0.42. Genomic BVs for RFI included 60,671 genetic markers for 1.4 million Holsteins and produced calculated genomic reliabilities for young animals averaging 19% compared with traditional reliabilities of 5%. A parallel test using somatic cell score (SCS) records for these same 3,947 cows indicated lower observed than expected genomic reliability (11% vs. 19%). The economic value of RFI is very large and could receive >20% of total emphasis in net merit, but the low reliability will limit the extra genetic progress to about 4% more than current progress. The RFI predictions were added to the extra feed associated with large BWC and then multiplied by −305 to convert from extra feed eaten per day to feed saved per lactation. Additional feed intake records could make feed saved a very important trait in future selection indexes for dairy cattle.