This paper gives an analysis of an attractor neural network model dubbed the Phonological Latching Network. The model appears to reproduce certain quintessentially phonological phenomena, despite not having any of these phonological behaviours programmed or taught to the model. Rather, assimilation, segmental-OCP, and sonority sequencing appear to emerge spontaneously from the combination of a few basic brain-like ingredients with a phonology-like feature system. The significance of this can be interpreted from two angles: firstly, the fact that the model spontaneously produces attested natural language patterns can be taken as evidence of the model’s neural and psychological plausibility; and secondly, it provides a potential explanation for why these patters appear to frequently in natural language grammars. Namely, they are a consequence of latching dynamics in the brain.