The objective of this study was to investigate if a multi breed reference would improve genotype imputation accuracy from 50K to high density (HD) single nucleotide polymorphism (SNP) marker data in Nordic Red Dairy Cattle, compared to using only a single breed reference, and to check the subsequent effect of the imputed HD data on the reliability of genomic prediction. HD genotype data was available for 247 Danish, 210 Swedish and 249 Finnish Red bulls, and for 546 Holstein bulls. A subset 50 of bulls from each of the Nordic Red populations was selected for validation. After quality control 612,615 SNPs on chromosome 1-29 remained for analysis. Validation was done by masking markers in true HD data and imputing them using Beagle v. 3.3 and a reference group of either national Red, combined Red or combined Red and Holstein bulls. Results show a decrease in allele error rate from 2.64, 1.39 and 0.87 percent to 1.75, 0.59 and 0.54 percent for respectively Danish, Swedish and Fi nnish Red when going from single national reference to a combined Red reference. The larger error rate in the Danish population was caused by a subgroup of 10 animals showing a large proportion of Holstein genetics, which was not present in the reference. When adding Holstein animals to the reference, the error rate was further decreased to 1.17 percent for Danish Red, whereas the Swedish and Finnish Red animals were unaffected. Three Danish animals with a high proportion of genetic origin from breeds not included in the reference were also unaffected by the addition of Holstein to the reference. Results show that a multi breed reference is superior to a single breed reference when imputing from 50K to HD marker data in the Nordic Red breeds, due to sharing the information of Holstein which was previously used in Danish Red. Investigation of the reliability of genomic prediction using imputed HD data shows small improvements compared to 50K data.