The complex morphotonology of the Tonga verb has offered a challenge to generative phonology for at least thirty years. In this paper I approach these data with new tools, the Optimal Domains theory of Cassimjee and Kisseberth (1997). I argue for an analysis incorporating (1) a tone reversal, i.e. treatment of historical toneless syllables as sponsors of H tone, (2) the separation of tone domain construction from the surface realization of tone, in the spirit of Optimal Domains theory, and (3) the presence of domain boundaries in the underlying representations. These assumptions lead to a relatively straightforward analysis of the tonal alternations.