In languages as diverse as English, Arabic, and Russian, the cooccurrence frequency of two consonants can be predicted from a gradient measure of the similarity of the phonological make-up of the two consonants. An argument for the role of phonological similarity in these languages is that it tends to be the case that sound classes with more members have weaker cooccurrence restrictions than classes with fewer members. More members require more contrasts, which reduces similarity. In Arabic, there are weaker restrictions on coronals (n=11) than labials (n=4) because coronals exhibit more contrasts. However, there are a number of well-documented Polynesian languages (Maori, Hawaiian, Tongan) were the reverse is true: there are more labial consonants than coronals, but there are stronger cooccurrence restrictions on labials than coronals. This project investigates the lexical statistics of these languages and proposes a revised characterization of similarity avoidance that has greater cross-linguistic coverage.