Abstract Phonaesthemes like the onset / sn- / in snore, snort, and sniffle straddle the boundary of phonology and morphology and thus seemingly argue against the “theoretical desideratum” of absolute modularity in the grammatical architecture (Lowe, 2016). If modularity is psychologically real, however, a “correction” of the phonaesthemic problem should occur so that phonaesthemes either become more clearly phonological or more clearly morphological over time. Since the morpheme-like nature of phonaesthemic onsets is grounded in their type frequency (Bergen, 2004), a development toward either pole should manifest itself in a change of this frequency. This hypothesis is investigated by collecting all roots beginning with four onset clusters of phonaesthemic value ( fl- , gl- , sl- , sn- ) in Middle and in Present-Day English. The comparison of these counts shows that there is a significant decrease of the phonaesthemic share in only one group ( sl- ), mostly due to loanwords. In the other groups, the ratio of phonaesthemic to arbitrary roots has remained virtually the same. I take these results to show that a strict separation between phonology and morphology is psychologically implausible. In order to account for the diachronic stability of phonaesthemes, I invoke word-based approaches to morphology, in which segmentable morphemes are not theoretical primitives.