1. DNA barcoding and metabarcoding are increasingly used as alternatives to traditional morphological identifications. For animals, the standard barcode is a c. 658-bp portion of the COI gene, for which reference libraries now cover a large proportion of described mammal species. Unfortunately, because its sequence is too long and does not contain highly conserved primer binding sites, this marker is not adapted for metabarcoding. Although alternative metabarcodes have been developed, their performances are generally seldom assessed. 2. We evaluate the reliability of a short metabarcode located in the mitochondrial 12S ribosomal RNA for the identifications of Amazonian mammals. We (i) constitute a nearly exhaustive reference library for species found in French Guiana, (ii) assess the taxonomic resolution of the marker and validate its use with dipteran blood meal analyses, (iii) assess the conservation of the primer binding sites, and (iv) compare its theoretical performances with that of a newly designed metabarcode located within the standard COI barcode. 3. About 576 specimens representing 164 species were gathered and sequenced. We show that the 12S marker allows remarkably accurate taxonomic assignations despite its very short size, and that primer binding sites are highly conserved, which is important to avoid PCR amplification bias potentially leading to detection failure. Additionally, our results stress that the identifications should only be considered at the generic level when they are based on incomplete reference libraries, even when a stringent similarity cutoff is used. A new short COI metabarcode was designed based on 569 reference sequences of mammals retrieved on BOLD. Our results clearly show that, while both markers provide similar taxonomic resolution, much higher rates of primer mismatches are found with COI. 4. Besides demonstrating the accuracy of the short 12S marker for the identification of Amazonian mammals and providing a reliable molecular reference database, this study emphasizes that the accuracy of taxonomic assignations highly depends on the comprehensiveness of the reference library and that great caution should be taken for interpreting metabarcoding results based on scarce reference libraries. The comparison with a short COI metabarcode also provides novel evidence in support for the use of ribosomal markers in metabarcoding studies.