The effect of the reference population size and the number of missing single nucleotide polymorphisms (SNPs) on imputation accuracy was determined. The population imputation method using the FImpute software was applied. The dataset used for the purpose of this study was taken from the database of the Holstein Cattle Breeders Association of the Czech Republic. It contains 1000 animals genotyped with the Illumina BovineSNP50 v.2 BeadChip. Two datasets were created, the first containing the original genotypes, including the missing SNPs, the second containing the same genotypes modified to avoid missing data. In these datasets, animals were randomly selected for a reference population (10, 25, 50 and 75%) and there were randomly selected SNPs for deletion (15, 30, 55, 70, and 95%) in animals that were not used as the reference population. Subsequently, the data accuracy was determined by two parameters: correlation between original and imputed SNPs and percentage of correctly imputed SNPs. Since animals and SNPs were randomly selected, the process including data imputation was repeated 100 times. Accuracy was determined as the average accuracy over all repetitions. It was found that the imputation accuracy is influenced by both parameters. If the size of the reference population is sufficient, the imputation accuracy is higher despite the large number of missing SNPs.