Genomic prediction has the potential to accelerate the genetic improvement rate for feed efficiency traits in beef cattle. In this study, we evaluated genomic prediction accuracies for residual feed intake (RFI) and its component traits dry matter intake (DMI), average daily gain (ADG), and metabolic body weight (MWT) based on genotyped 50K and imputed 7.8 million whole genome sequence SNPs in multiple Canadian beef cattle populations. The populations included purebred Angus (N=1,162), purebred Charolais (N=717), Kinsella (N=1,506), Elora (N=775), PG1 (N=1,911), and TX (N=1,502). Animals from the six populations were combined into a single reference population and genomic prediction was conducted using GBLUP based on 50K (50K-GBLUP) and 7.8 million imputed SNPs (seqGBLUP) with 5-fold cross validation of each population. In addition, a weighted GBLUP (w-seqGBLUP) was performed for the 7.8 million imputed SNPs using a G matrix constructed by weighting SNPs of nine functional classes with their weighting factors obtained based on the average square of estimated marker effects of each functional class from GWAS. The results showed that both seqGBLUP and w-seqGBLUP yielded similar accuracies for all the traits of all breed populations. For crossbred populations, seqGBLUP and w-seqGBLUP improved the prediction accuracy by 4.1%, or from the realized genomic prediction accuracy of 0.363 to 0.378 for RFI of the Kinsella population, to 16.4% or from 0.311 to 0.362 for ADG of the Elora population in comparison to the 50K-GBLUP. However, both seqGBLUP and w-seqGBLUP had a 6.6% to 11.6% lower prediction accuracy than that of 50K-GBLUP for purebred Angus. A reduction of 1.3% and 1.5% on genomic prediction accuracy was also observed for MWT and RFI, respectively, for purebred Charolais. On-going studies are being undertaken to further improve genomic prediction accuracies for feed efficiency traits in Canadian beef cattle.