The analysis of the skeletal remains of vertebrates in archaeological contexts provides information about human-animal relationship and their environment. Their taxonomic identification based on macroscopic observation is not always possible due to fragmentation and poor preservation. In recent years, proteomics has emerged as an alternative but there is clearly a lack of data in arid environment where diagenesis rapidly affects the integrity of bone proteins. Here, we report the efficiency of three protocols for protein extraction. The protocols used harsh (1 M HCl and 0.6 M HCl) and soft (Tris-EDTA) decalcification agents and were tested on unidentified splinters from the 2000 years-old site of Toteng, Botswana. The preservation of the organic phase was first estimated using attenuated total reflectance Fourier transform infrared spectroscopy and a set of samples with contrasted collagen contents were selected for palaeoproteomics. The extracted proteins were submitted to a bottom-up proteomic approach involving trypsin digestion followed by ultra-high-performance liquid chromatography coupled to mass spectrometry (UHPLC-MS). Our results identify Tris-EDTA buffer as the most suitable decalcification protocol for poorly preserved bones and propose a collagen content threshold of ~3% weight content for successful detection of peptides. This approach, combined with biogeographical and chronological repartitions of mammals in Africa allows refining taxonomic attributions for four out of nine splinters, leading to species identification. Data are available via ProteomeXchange with identifier PXD010725.