In the earliest work on tone languages, tones were treated as atomic units: High, Mid, Low, High Rising, etc. Universal tone features were introduced into phonological theory by Wang 1967 by analogy to the universal features commonly used in segmental phonology. The implicit claim was that features served the same functions in tonal phonology as in segmental phonology. However, with the advent of autosegmental phonology (Goldsmith 1976), much of the original motivation for tone features disappeared. Contour tones in many languages were reanalyzed as sequences of simple level tones, calling into question the need for tonal features such as [±falling]. Processes of tone copy such as L(ow) > H(igh) / __ H(igh) were reinterpreted as tone spreading instead of feature assimilation. At about the same time, a better understanding of downstep emerged which allowed many spurious tone levels to be eliminated. As a result, in spite of the vast amount of work on tone languages over the past thirty years, the number of phenomena that appear to require tone features has become significantly reduced, raising the issue whether the notion of tone features is at all useful. This paper first reviews the basic functions for which segmental features have been proposed, and then examines the evidence that tone features are needed to serve these or other functions in tone languages. The discussion focuses successively on level tones, contour tones, and register, building on examples from Africa and Asia. Our current evaluation of the evidence is that tone features, to the extent that they appear motivated at all, do not serve the same functions as segmental features.