Log Size Index (LSI) has been developed and used in archaeozoology to perform biometric studies of often small datasets, a recurrent limitation of archaeological faunal assemblages. This approach consists of aggregating measurements taken on different anatomical parts by standardisation, using a reference to which the studied measurements are scaled. No consensus exists in the choice of the reference to be used for LSI analyses and several datasets are now available, differing in their specimens and variables composition, and geographic and temporal origin. Using a biometric dataset of 5533 measurements of cattle, suids, sheep and goat from southern France and dated from the Roman times we compared the use of several references, provided by the R package zoolog, with a ‘mean’ reference specifically built for our studied sample and that corresponds to an averaged specimen. We demonstrated that references differ both in their raw measurements and their allometries, and that the choice of the reference influences the number of variables and measurements that can be analysed, but also diachronic LSI analyses and potentially their interpretations. While the main patterns are similar among references, local variation exist and it could not be excluded that the choice of the reference will more strongly influence other studies. We advocate to use the mean of the studied archaeological assemblage as reference, with the double benefit of exploiting to the maximum the number of available measurements, and of minimising, on average, the possible allometric differences between the studied population and the reference.