Loanwords are one of the main sources of information for the study of in- teractions in ancient and prehistoric contexts. Hence, distinguishing potential loanwords from inherited words is a crucial and often controversial issue. In the past, scholars have deployed language-specific approaches (e.g. looking for phonotactic inconsistencies – cf. Kang 2011), but more recently, the rise of new computational and phylogenetic methods has led to a growing interest in the possibility to identify features and patterns that characterize loanwords cross-linguistically (Zhang, Fabri, and Nerbonne 2021; Miller et al. 2020; Nath et al. 2022). This contribution belongs to this second line of research. It stems from the anecdotal observation that, within a language, loanwords of- ten seem to pick from poorer phonemic inventories than inherited words. In order to investigate this intuition, we first performed a series of statistical anal- yses on the WOLD database (Haspelmath and Tadmor 2009; 2021). In particular, we verified that loanwords tend to contain fewer rare and/or co- articulated phonemes (e.g. labial-velars and ejectives). Since the statistical data yielded positive results, we then built a model verifying two hypotheses for a partial explanation of this phenomenon: 1) loanwords tend to contain fewer cross-linguistically rare phonemes than inherited words simply because being cross-linguistically rare, they are less likely to be found both in the source and in the target languages, and hence more likely to be reinterpreted and replaced by more common phonemes when borrowed; 2) loanwords tend to contain fewer co-articulated phonemes because phonemes of the source words that are absent in the target languages tend to be replaced by mono-articulated phonemes more often than by co-articulated phonemes. This could be explained by the hypoth- esis that co-articulated phonemes are more often perceived as marked, and that phoneme replacements tend to favor less marked phonemes. As will be discussed in the conclusions, the results we have obtained raise interesting questions and problems that are relevant for the search of loanwords in ancient/prehistoric contexts. Finally, we will also mention a few of the practical issues we have encountered in our analyses, as they highlight gaps in the scholarship and point to potential avenues for further research.