The African leopard (Panthera pardus pardus) has lost a significant proportion of its historical range, notably in north-western Africa and South Africa. Recent studies have explored the genetic diversity and population structure of African leopards across the continent. A notable genetic observation is the presence of two divergent mitochondrial lineages, PAR-I and PAR-II. Both lineages appeared to be distributed widely, with PAR-II frequently found in southern Africa. Until now, no study has attempted to date the emergence of either lineage, assess haplotype distribution, or explore their evolutionary histories in any detail. To investigate these underappreciated questions, we compiled the largest and most geographically representative leopard data set of the mitochondrial NADH-5 gene to date. We combined samples (n = 33) collected in an altitudinal transect across the Mpumalanga province of South Africa, where two populations of leopard are known to be in genetic contact, with previously published sequences of African leopard (n = 211). We estimate that the maternal PAR-I and PAR-II lineages diverged approximately 0.7051 (0.4477–0.9632) million years ago (Ma). Through spatial and demographic analyses, we show that while PAR-I underwent a mid-Pleistocene population expansion resulting in several closely related haplotypes with little geographic structure across much of its range, PAR-II remained at constant size and may even have declined slightly in the last 0.1 Ma. The higher genetic drift experienced within PAR-II drove a greater degree of structure with little haplotype sharing and unique haplotypes in central Africa, the Cape, KwaZulu-Natal and the South African Highveld. The phylogeographic structure of PAR-II, with its increasing frequency southward and its exclusive occurrence in south-eastern South Africa, suggests that this lineage may have been isolated in South Africa during the mid-Pleistocene. This hypothesis is supported by historical changes in paleoclimate that promoted intense aridification around the Limpopo Basin between 1.0–0.6 Ma, potentially reducing gene flow and promoting genetic drift. Interestingly, we ascertained that the two nuclear DNA populations identified by a previous study as East and West Mpumalanga correspond to PAR-I and PAR-II, respectively, and that they have come into secondary contact in the Lowveld region of South Africa. Our results suggest a subdivision of African leopard mtDNA into two clades, with one occurring almost exclusively in South Africa, and we identify the potential environmental drivers of this observed structure. We caution that our results are based on a single mtDNA locus, but it nevertheless provides a hypothesis that can be further tested with a dense sample of nuclear DNA data, preferably whole genomes. If our interpretation holds true, it would provide the first genetic explanation for the smaller observed size of leopards at the southernmost end of their range in Africa.