We tested the reliability of herbivore faecal d13C and d15N values for reconstructing diet through review of an extensive database derived from a 3-year study of ungulates in South Africa’s Kruger National Park. Faeces are a useful material for stable isotope studies of diet because they record dietary turnover at very short time scales, and because sampling is non-invasive. However, the validity of faecal isotope proxies may be questioned because they represent only undigested food remains. Results from Kruger Park confirm that free-ranging browsers have faecal d13C consistent with C3 feeding, grazer faeces are C4, and mixed-feeder faeces intermediate. Although the respective ranges do not overlap, there is significant variation in faecal d13C of browsers and grazers (2.0–4.0%) across space and through time. We demonstrate that most (70%) of this variation can be ascribed to corresponding patterns of variation in the d13C of C3 and C4 plants, respectively, re-enforcing the fidelity of faecal isotope proxies for diet but highlighting a need for mixing models that control for variations in plant d13C in order to achieve accurate diet reconstructions. Predictions for the effects of climate (rainfall) and ecophysiology on 15N-abundance variations in mammals do not persist in faeces. Rather, faecal d15N tracks changes in plant d15N, with further fractionation occurring primarily due to variations in dietary protein (reflected by %N). Controlling for these effects, we show that a dual-isotope multiple source mixing model (Isosource) can extend diet reconstructions for African savanna herbivores beyond simplified C3/C4 distinctions, although further understanding of variations in mammal d15N are needed for greater confidence in this approach.