The antipassive, an object-demoting diathesis commonly associated with ergative languages (Cooreman 1994; Dixon 1994; Kulikov 2011; among others), has so far largely gone unnoticed in Bantu languages, which are of the accusative type. Although its occurrence in accusative languages has received recent attention (Janic 2013), this object-demoting diathesis is commonly believed not to exist in Bantu. The WALS map presented by Polinsky (2011) is telling in this regard: the only six Bantu languages covered are all marked “no antipassive”. In this paper, we present comparative evidence to demonstrate that the antipassive is a voice construction to be reckoned with in Bantu. A robust typology of Bantu voice constructions has been developed on the basis of the scarce data available in the literature (Bostoen et al. 2015). The evidence stems from a sample of nearly 200 studies covering 165 different languages, and is furthermore reinterpreted in the light of original data from a number of Bantu languages, such as Cilubà (Dom et al. 2015) and Kirundi (Ndayiragije 2006), which were the subject of a more in-depth analysis. It is shown how the antipassive generally developed as a specific reading of the highly polysemous verbal suffix -an-, which is more commonly used as a reciprocal/associative marker (Dammann 1954; Meeussen 1967; Schadeberg 2003). These and other functions can be accounted for by the underlying notion of “plurality of relations” (Lichtenberk 2000: 34), which is characterized by a low degree of participant/event elaboration. From both a diachronic and syntactic point of view, it is argued that the development of antipassives out of plurality constructions results from the gradual demotion of the second participant of a co-participative event.