Ancient DNA preservation in subfossil specimens provides a unique opportunity to retrieve genetic information from the past. As ancient DNA extracts are generally dominated by molecules originating from environmental microbes, capture techniques are often used to economically retrieve orthologous sequence data at the population scale. Post-mortem DNA damage, especially the deamination of cytosine residues into uracils, also considerably inflates sequence error rates unless ancient DNA extracts are treated with the USER enzymatic mix prior to library construction. While both approaches have recently gained popularity in ancient DNA research, the impact of USER-treatment on capture efficacy still remains untested. In this study, we applied hyRAD capture to eight ancient equine subfossil specimens from France (1st-17th century CE), including horses, donkeys and their first-generation mule hybrids. We found that USER-treatment could reduce capture efficacy and introduce significant experimental bias. It differentially affected the size distribution of on-target templates following capture with two distinct hyRAD probe sets in a manner that was not driven by differences in probe sizes and DNA methylation levels. Finally, we recovered unbalanced proportions of donkey-specific and horse-specific alleles in mule capture sequence data, due to the combined effects of USER-treatment, probe sets and reference bias. Our work demonstrates that while USER-treatment can improve the quality of ancient DNA sequence data, it can also significantly affect hyRAD capture outcomes, introducing bias in the sequence data that is difficult to predict based on simple molecular probe features. Such technical batch effects may prove easier to model and correct for using capture with synthetic probes of controlled sizes and diversity content.