The issue of vowel height harmony – relatively rare in the world's languages – is one that most serious theories of phonology have addressed at one time or another, particularly as concerns its realisation in Bantu (e.g. Clements 1991, Archangeli & Pulleyblank 1994, Beckman 1997). As is quite well known, the majority of an estimated 500 Bantu languages exhibit some variant of a progressive harmony process by which vowels lower when preceded by an appropriate (lower) trigger. (Ki)-Yaka, a Western Bantu language spoken in ex-Zaire, designated as H.31 by Guthrie (1967–71), has a height harmony system which has been analysed as having a similar left-to-right lowering process. In this paper I argue against the general analysis given for Yaka, showing that this language differs in a major way from the rest of Bantu. The goals of the paper are threefold. First, I present a comprehensive treatment of the unusual vowel harmony system in (ki-)Yaka. Second, I introduce the notion of the ‘prosodic trough’ (τ), a domain which is needed in order to state important phonological generalisations in Yaka and in Bantu in general. Finally, I show the relevance of the Yaka facts for the study of positional prominence in phonology. A (partial) analysis is offered within optimality- theoretic terms, particularly as developed by McCarthy & Prince (1995). Although superficially resembling the vowel height harmony found in most Bantu languages, the Yaka system will be shown to differ from these latter in major ways. The paper is organised as follows. In §2 I establish the general nature of the Yaka harmony system, reanalysing previous accounts in terms of ‘plateauing’. In §3 I turn to the process of ‘imbrication’, which introduces a second motivation for vowel harmony: the avoidance of the sequence [wi]. A third source of vowel harmony is presented in §4, which also introduces the notion of the ‘prosodic trough’. The study ends with a brief conclusion in §5 and an appendix that discusses outstanding problems.