Rather than having a binary H vs. L tone system like the other members of the Dogon family are purported to have, Tommo So demonstrates a ternary distinction of H vs. L vs. underspecified. Underspecified tones are grammatically constrained to clitics, certain suffixes, and epenthetic vowels; crucially, they may not surface on any lexical stems unless they are derived there by the addition of one of the grammatical elements listed before. I argue that this distribution fits naturally into our typology of stress and grammatical elements. Underspecification in Tommo So is not simply a phonological tool meant to explain the distribution of fully specified surface tones. Instead, it persists to the surface, with the F0 values of underspecified syllables filled in by linear interpolation between surrounding specified points, yielding non-categorical F0 contours interspersed with canonical Hs, Ls, and true contour tones. Chichewa (Myers 1998) shows a similar system of surface underspecification, but with only one specified tone, H. The existence of these cases shows the importance of doing instrumental analysis when describing a tone system.