In Shilluk (Western Nilotic, South Sudan), a tonal suffixation analysis of nominal inflection is challenged by the fact that there is no clear morphological base for nouns and that grammatical tone effects are not always constant for a given category. I show that both of these problems for the concatenative analysis disappear if one decomposes tone into an internally structured set of subtonal features, and that tones may be underspecified in various ways.