Abstract Within evolutionary biology, mitochondrial genomes (mitogenomes) provide useful insights at both population and species level. Several approaches are available to assemble mitogenomes. However, most are not suitable for divergent, extinct species, due to the requirement of a reference mitogenome from a conspecific or close relative, and relatively high-quality DNA. Iterative mapping can overcome the lack of a close reference sequence, and has been applied to an array of extinct species. Despite its widespread use, the accuracy of the reconstructed assemblies are yet to be comprehensively assessed. Here, we investigated the influence of mapping software (BWA or MITObim), parameters, and bait reference phylogenetic distance on the accuracy of the reconstructed assembly using two simulated datasets: (i) spotted hyena and various mammalian bait references, and (ii) southern cassowary and various avian bait references. Specifically, we assessed the accuracy of results through pairwise distance (PWD) to the reference conspecific mitogenome, number of incorrectly inserted base pairs (bp), and total length of the reconstructed assembly. We found large discrepancies in the accuracy of reconstructed assemblies using different mapping software, parameters, and bait references. PWD to the reference conspecific mitogenome, which reflected the level of incorrect base calls, was consistently higher with BWA than MITObim. The same was observed for the number of incorrectly inserted bp. In contrast, the total sequence length was lower. Overall, the most accurate results were obtained with MITObim using mismatch values of 3 or 5, and the phylogenetically closest bait reference sequence. Accuracy could be further improved by combining results from multiple bait references. We present the first comprehensive investigation of how mapping software, parameters, and bait reference influence mitogenome reconstruction from ancient DNA through iterative mapping. Our study provides information on how mitogenomes are best reconstructed from divergent, short-read data. By obtaining the most accurate reconstruction possible, one can be more confident as to the reliability of downstream analyses, and the evolutionary inferences made from them.