Although runoff is usually quantified as a fraction of annual rainfall, its spatial heterogeneity and temporal variability are critical aspects poorly assessed in flat drylands. In this study, we: (i) quantified the annual runoff and its relation with watershed and piosphere (high cattle impact) areas, (ii) quantified the uneven contribution of the largest rainfall and runoff events to their respective yearly totals at different temporal scales, (iii) evaluated the runoff responses to rainfall characteristics (event size and intensity), and (iv) quantified the lag time between rainfall and runoff peaks and its relation to the maximum length of the watershed. To do this, we combined high-resolution field measurements (three consecutive years, 15-min resolution) with remote-sensing analyses for six watersheds (sizes between 20 and 2000 ha) in Dry Chaco rangelands (Argentina). On average, runoff was not explained by watershed or piosphere area. Runoff was more variable than rainfall: the 10 wettest days explained 100% and 60% of annual runoff and rainfall, respectively. Only 15% of the rainfall events generated runoff. Runoff required on average the coincidence of a 20.3 +/- 1.6 mm rainfall size and 12.2 +/- 2.2 mm h(-1) rainfall intensity. Above these thresholds, the response between rainfall and runoff was linear (R-2 > 0.66 and 0.62 for event size and intensity, respectively). Lag was 43 +/- 27 min, highlighting the ephemeral flash-flood nature of runoff. There was also a linear relationship between mean lag and watershed's maximum length (r = 0.87). Our results highlight the stochastic nature of runoff for Dry Chaco rangelands, both in space and time, modulated by rather subtle watershed physical elements and microtopography. This view challenges the idea of a fixed runoff generation area, which needs to be replaced by a more flexible, if not idiosyncratic concept, based on the interactions between terrain features and rainfall event characteristics.