This paper presents new evidence supporting the inclusion of the colon – a constituent intermediate between the foot and the prosodic word and made of two feet – as a separate category in the Prosodic Hierarchy. Many authors have used the colon to account for tertiary stress (e.g., Stowell 1979; Halle and Clements 1983:18-19; Hammond 1987; Green 1997; a.o.). However, most recent work in metrical theory rejects this category, arguing that it is unnecessary and too costly (e.g., Ellenbaas and Kager 1999; Hyde 2002; Martínez-Paricio and Kager 2015, 2016; a.o.). I analyze an unusual downstep pattern in Paicî (Oceanic) and show that the most satisfying account involves reference to the colon. Paicî is a two-tone language (H vs. L), where L-toned words of 4+ morae undergo register lowering after the second mora. This downstep does not occur with shorter words. I show that the two initial morae of 4+ mora words are parsed into a bimoraic foot. Crucially, no foot is parsed in 2 mora or 3 mora words. I argue that foot-parsing can only occur if it creates feet that are licensed by a colon (hence the need for at least 4 morae), making the Paicî downstep pattern the first piece of tonal evidence in favor of the colon. Alternatives using only traditional prosodic categories and mechanisms are shown to be more stipulative and less explanatory.