Vowel harmony is extremely widespread among the Bantu languages (see e.g. Clements 1991; Hyman 1999; Odden 2015). In this thesis, I present an investigation of progressive vowel height harmony in five-vowel Bantu languages using quantitative and experimental data, the results of which contribute to our understanding of the grounding of phonological patterns and have implications for synchronic formal analyses. A defining aspect of height harmony in Bantu is the asymmetric behaviour of front and back vowels. Though this is characteristic of the "canonical" variety of height harmony (after Hyman 1999), a front-back asymmetry is also seen in many non-canonical systems of height harmony. Certain non-canonical languages can even be seen to lack front height harmony but not back height harmony. To explore this and further issues, I present a study of vowel-pair frequencies in canonical Chewa, Kalanga and Yao and non-canonical Pende, Lozi and Makhuwa in which alternations due to height harmony are found only with verbal extensions. Here, I concentrate on vowel pairs in nouns. In general, pairs considered non-harmonic in height harmony were not necessarily under-represented in nouns, though this was the case in specific instances. For example, both [e.i] and [o.u] are under-represented in nouns; however, [o.u] was consistently more under-represented than [e.i]. This is reminiscent of the typological observation that though certain Bantu languages exhibit only back height harmony, none are known to possesses only front height harmony (Hyman 1999: 245). Together, these facts suggest that the avoidance of these pairs is particularly well motivated but more so for [o.u] than [e.i]. In addition, I argue that, upon closer examination, though alternations are only seen in verbs, Lozi in fact shows a wider-ranging phonotactic prohibition against [o.u] alone and that other gaps in verbs are synchronically accidental rather than phonotactic. This is followed by a production experiment carried out with speakers of canonical Bemba and Nyanja and non-canonical Lozi which investigated the potential effects of vowel-to-vowel coarticulation on F1 in relation to harmony. The results show that, overall, differences in F1 due to coarticulation do not echo the alternations of height harmony. In the case of [e.i] and [o.u], the production experiment does not fully align with the frequency study. Though there was evidence for lowering of [i] after [e], the evidence for lowering of [u] after [o] was weaker, thus showing an asymmetry in the opposite direction to both the typology and findings of the frequency study. The combined results of these studies may be seen either as suggesting that the grounding for progressive height harmony is perceptual rather than articulatory or that the coarticulatory processes that could have given rise to harmony belong to the phonetics of a previous stage of such Bantu languages and that this has since changed. Additionally, the results show how it can be informative to employ a whole language approach when analysing a language's phonology as this can provide further details not found in data pertaining only to a particular process which are key to the system as a whole.