Specimens stored in museum collections represent a crucial source of morphological and genetic information, notably for taxonomically problematic groups and extinct taxa. Although fluid-preserved specimens of groups such as teleosts may constitute an almost infinite source of DNA, few ancient DNA protocols have been applied to such material. In this study, we describe a non-invasive Guanidine-based (GuSCN) ancient DNA extraction protocol adapted to fluid-preserved spec-imens that we use to re-assess the systematics of the genus Orestias (Cyprinodontidae: Teleostei). The latter regroups pup-fishes endemic to the inter-Andean basin that have been considered as a ‘species flock’, and for which the morphology-based taxonomic delimitations have been hotly debated. We extracted DNA from the type specimens of Orestias kept at the Muséum National d’Histoire Naturelle of Paris, France, including the extinct species O. cuvieri. We then built the first molecular (control region [CR] and rhodopsin [RH]) phylogeny including historical and recently collected representatives of all the Orestias complexes as recognized by Parenti (1984a): agassizii, cuvieri, gilsoni and mulleri. Our ancient DNA extraction protocol was validated after PCR amplification through an approach based on fragment-by-fragment chimera detection. After optimization, we were able to amplify < 200 bp fragments from both mitochondrial and nuclear DNA (CR and RH, respectively) from probably formalin-fixed type specimens bathed entirely in the extraction fluid. Most of the individuals exhibited few modifications of their external structures after GuSCN bath. Our approach combining type ma-terial and ‘fresh’ specimens allowed us to taxonomically delineate four clades recovered from the well-resolved CR tree into four redefined complexes: agassizii (sensu stricto, i.e. excluding luteus-like species), luteus, cuvieri and gilsoni. The mulleri complex is polyphyletic. Our phylogenetic analyses based on both mitochondrial and nuclear DNA revealed a main, deep dichotomy within the genus Orestias, separating the agassizii complex from a clade grouped under shallow dichotomies as (luteus, (cuvieri, gilsoni)). This ‘deep and shallow’ diversification pattern could fit within a scenario of ancient divergence between the agassizii complex and the rest of Orestias, followed by a recent diversification or adaptive radiation within each complex during the Pleistocene, in- and outside the Lake Titicaca. We could not recover the recip-rocal monophyly of any of the 15 species or morphotypes that were considered in our analyses, possibly due to incomplete lineage sorting and/or hybridization events. As a consequence, our results starkly question the delineation of a series of diagnostic characters listed in the literature for Orestias. Although not included in our phylogenetic analysis, the syntype of O. jussiei could not be assigned to the agassizii complex as newly defined. The CR sequence of the extinct O. cuvieri was recovered within the cuvieri clade (same haplotype as one representative of O. pentlandii), so the mtDNA of the former species might still be represented in the wild.