The objective of the study was to estimate the efficiency of imputation of genotypes of the Affymetrix Axiom Equine genotyping array, including 670,806 SNP, from genotypes of the Illumina Equine SNP50 BeadChip including 54,602 SNP and the Illumina Equine SNP-74k-chip including 65,157 SNP. Genotypes were avalaible from 5 breeds: Arabs (AR, 1,207 horses), Selle Français (SF, 1,978 horses), Trotteur Français (TF, 979 horses), Anglo-Arabs (AA, 229 horses), and various foreign sport horses (FH, 209 horses). In AR, 15% of horses were genotyped with the high density chip, in SF 57%, in TF 30%, in AA 10% and FH 15%. A validation set equal to the third of the sample was drawn in horses genotyped with high density chip and their genotypes blinded. Sampling was made to mimic the parental structure of the target population (the one genotyped with medium density), i.e. at random in SF and AR and with a genotyped sire for TF. Two strategies were compared with a reference population of all breeds and with only the same breed as the validation set. The software FImpute was used. First results in TF showed better concordance with a validation set of the same breed: mean of 98.0% of correctly imputed genotypes (s.d. 2.3%) versus 96.0% (s.d. 5.6%) with a validation set of all breeds. Theses results were compatible with the use of imputed data for future GWAS and genomic evaluation for these data.