Studies of segmental co-occurrence constraints have consistently produced evidence consistent with a cross-linguistic anti-similarity bias for consonants (e.g. Frisch et al. (2004), Pozdniakov & Segerer (2007), Walter (2010), Doucette et al. (2024)). The current study tests whether, in spite of this universal similarity avoidance bias, there are cross-linguistically consistent featural harmony biases the world’s lexicons. In particular, we test for the presence of a nasal consonant harmony bias, given the fact that categorical nasal consonant harmony is attested in multiple language families and that nasal harmony is phonetically-motivated. A Bayesian negative binomial model of 91 typologically diverse languages’ type frequencies for two-consonant words shows evidence of a weak but reliable cross-linguistic bias in favor of nasal harmony, as well as a comparable bias in favor of voicing harmony. The findings also show patterns consistent with a similarityavoidance bias, most notably a strong cross-linguistic bias against coronal harmony. Taken together, these findings support the notion that similarity-based co-occurrence constraints may be feature dependent in cross-linguistically consistent ways, and more generally that featural dimensions are relevant for understanding the role of segmental redundancy in lexicons.