Water repellent soils represent some of the most responsive natural surfaces from the point of view of overland flow generation. Measurement of the runoff response of such soils can therefore provide characteristics that can help constrain hydrological models or improve the conceptual models of hillslope runoff generation. We have analysed the overland flow characteristics of water repellent sandy soils at three sites with different dominant vegetation (hereafter denoted as the lichen, moss and herb sites) and following vegetation removal (scalping) during extreme rainfall events created by a rainfall simulator (three consecutive sprinklings at 2.1 mm min−1 for 15 min each with a 5 min hiatus between the sprinklings). Despite the extreme water repellency (WR) and rainfall intensity, on average just up to a half of the rainfall left the plots as an overland flow. Whilst the overland flow mostly generated quickly (on average within 50–120 s), the vegetation substantially delayed its appearance (by approximately 100% compared to scalped plots). The median values of the runoff coefficients were 6% and 49% on the soil with and without vegetation, respectively. Good relationships between the overland flow characteristics and soil moisture were found only for the scalped plots. Although the time to runoff and runoff coefficients after each of the three consecutive sprinklings were not substantially different, the differences between the soil with and without vegetation remained substantial. The results indicated: (a) high variability within each of the sites, including the lichen site that appeared homogenous; (b) all vegetated plots retained a higher amount of water than the scalped plots; (c) although all sites exhibited high WR, the combined influence of preferential flow, higher hydraulic conductivity and specific vegetation (soapwort, black locust) mitigated its effects at the herb site; (d) the relatively lower overland flow at that site may point to subsurface flow.