Technical Abstract: The objective of this study is to investigate single nucleotide polymorphism (SNP) genotypes imputation of Hereford cattle. Purebred Herefords were from two sources, Line 1 Hereford (N=240) and representatives of Industry Herefords (N=311). Using different reference panels of 62 and 494 males with 1954 SNPs on Bos Taurus autosome 6, and a fixed study group of 57 Line 1 females for which genotypes were available for 10% of loci, imputation were performed to predict the other 90% loci. Imputation algorithm DAGPHASE (Tom Druet; Unit of Animal Genomics, University of Liege, Liege, Belgium) relies on linkage disequilibrium and linkage. Reference panels consisted of subsets or all males from Line 1 Hereford, Industry Herford or both. Subsets from Industry Hereford were chosen randomly or based on relationship to study group using either pedigree derived family relationship or SNPs derived population structure. Average imputation accuracy went up when the size of reference panel increased. It reached a plateau when the ratio of reference panel size to study group size was larger than 4. The mean proportion of genotypes imputed correctly ranged from 0.868 to 0.923 when using reference panels from Line 1 Hereford, 0.745 to 0.823 when using reference panels of random samples from Industry Hereford, 0.802 to 0.859 when using highly related samples from Industry Hereford, 0.625 to 0.770 when using lowly related samples from Industry Hereford. Highly related samples based on SNPs derived population structure had 1.5 to 3.9% higher average imputation accuracy than highly related group samples based on pedigree derived family relationship. Our study implied reference panel size, family relationship and population structure can affect imputation accuracy.