This article revives old descriptive data on Awa, a Papuan language of the Kainantu group. The tonal system was described in detail in a paper by Loving (1973), where he reports a series of toneless noun suffixes, falling into six classes depending on their tonal alternations when combined with a noun root. This article demonstrates that the suffixes are best understood as carrying lexical tone; the alternations in form arise from the interaction of typologically natural tonotactic constraints. While the system can be described in autosegmental terms without much difficulty, a formal constraint-based analysis is less straightforward. I show that strict ranking, as in optimality theory (Prince & Smolensky 2004 ), fails to capture the data patterns due to cumulativity effects, some of which cannot be naturally captured even with local constraint conjunction (Smolensky 2006). The data are successfully modeled in harmonic grammar (Legendre et al. 1990).