We studied the degree and scale of patchiness of vegetation and selected soil variables along a gradient of herbivore impact. The gradient consisted of a radial pattern of `high', `intermediate' and `low' herbivore impact around a watering point in a semi-arid environment in Burkina Faso (West Africa). We hypothesised that, at a certain range of herbivore impact, vegetated patches alternating with patches of bare soil would occur as a consequence of plant-soil feedbacks and run-off-run-on patterns. Indeed, our transect data collected along the gradient showed that vegetated patches with a scale of about 5–10 m, alternating with bare soil, occurred at intermediate herbivore impact. When analysing the data from the experimental sites along the gradient, however, we also found a high degree of patchiness of vegetation and soil variables in case of low and high herbivore impact. For low herbivore impact, most variation was spatially explained, up to 100% for vegetation biomass and soil temperature, with a patch scale of about 0.50 m. This was due to the presence of perennial grass tufts of Cymbopogon schoenanthus. Patterns of soil organic matter and NH4-N were highly correlated with these patterns of biomass and soil temperature, up to r=0.7 (P<0.05) for the in situ correlation between biomass and NH4-N. For high herbivore impact, we also found that most variation was spatially explained, up to 100% for biomass and soil temperature, and 84% for soil moisture, with three distinct scales of patchiness (about 0.50 m, 1.80 m and 2.80 m). Here, microrelief had a corresponding patchy structure. For intermediate herbivore impact, again most variation was spatially explained, up to 100% for biomass and soil temperature, and 84% for soil moisture, with a patch scale of about 0.95 m. Here, we found evidence that vegetated patches positively affected soil moisture through less run-off and higher infiltration of rainwater that could not infiltrate into the bare soil elsewhere, which was not due to microrelief. Thus, we conclude that our findings are in line with our initial hypothesis that, at intermediate herbivore impact, vegetated patches alternating with patches of bare soil persist in time due to positive plant-soil feedbacks.