This article analyzes the non-lexical conversational sounds (conversational grunts) of English, including such items as uh-huh, un-hn, um, mm, and oh, based primarily on examination of a few hundred occurrences in a corpus of conversations. The data includes extensive phonetic variation, suggesting that these items are best explained, not as fixed words, but as dynamic creations. In particular, the vast majority of these items can be generated by a simple model consisting of 10 component sounds and 2 combining rules. Moreover, each of these component sounds seems to bear some meaning or function which is fairly constant across grunts and across contexts; and so the meanings of conversational grunts are largely a product of sound symbolism. This analysis is compatible with acoustic and cognitive factors which are present in the conversational contexts where grunts occur.