Sesotho is a tonal Southern Bantu language which has so far received extremely little attention by the speech research community. We consider tone modelling for Sesotho using the Fujisaki model-based analysis with a view to the development of a text-to-speech (TTS) system. Fujisaki analysis can be used to indicate the tone associated with a syllable, but it often differs from the surface tone that would be available for TTS synthesis. We investigate instances in which the surface tone differs from the tone indicated by Fujisaki analysis, and determine the effect of these discrepancies on speech quality. The amplitude of Fujisaki tone commands is manipulated to match the surface tones, and the resulting resynthesized speech subsequently analysed by perceptual tests. We find that the effect of inserting tone commands at high surface tone syllables is more severe than matching the Fujisaki tone commands with low surface tone syllables, in terms of naturalness. Furthermore, some discrepancies can be attributed to errors in the surface tonal transcription. However, on average, all manipulations lead only to a mild degradation in speech quality. We conclude that the Fujisaki model is a feasible way to model tone in Sesotho even in the presence of limited and under-developed linguistic resources.