Single-nucleotide polymorphism (SNP) arrays are commonly used in livestock genetics to investigate complex traits, including genome- wide associations and fine mapping, genomic prediction, and genetic diversity analyses. In a European equine diversity study, we analysed the Axiom Equine 670K SNP genotype data from 2,768 equids representing 20 horse breeds and one donkey breed. Using a strict detection setting to identify genome-wide runs of homozygosity (ROH), 169 purebred horses displayed fewer ROH segments than F1 crosses. Under medium and relaxed settings, ROH counts increased, while some horses continued to exhibit low ROH levels. Therefore, we additionally assessed genotype performance using a four-fold concordance analysis of replicate pairs on the same Axiom batch, between two different Axiom batches, between Illumina EquineSNP50 BeadChip and between Illumina paired-end HiSeq 2000 whole genome sequencing data. Replicates within the same Axiom batch showed the highest average genotype concordance (98.81%), followed by Illumina 50K (97.88%) and whole genome sequencing (96.84%). Re-genotyped horses with few ROH segments showed the lowest concordance (93.52%). According to the SNPolisher classification, 120,838 genome-wide SNPs were not recommended for reproducibility. After calling genotypes of the two different batches together following Axiom Best Practice (e.g. removing failing samples before the final genotyping) and excluding non-recommended SNPs, concordance improved in all comparisons. Therefore, we recommend excluding horses exhibiting an unusually high number of heterozygous calls, using only SNPs with validated genotype performance, and accounting for batch effects when analyzing Axiom Equine 670K SNP genotype data from different batches.