There has been limited success in determining critical thresholds of ground cover or soil characteristics that relate to significant changes in runoff or sediment production at the microscale (<1 m2), particularly in semi-arid systems where management of ground cover is critical. Despite this lack of quantified thresholds, there is an increasing research focus on the two-phase mosaic of vegetation patches and inter-patches in semi-arid systems. In order to quantify ground cover and soil related thresholds for runoff and sediment production, we used a data mining technique known as conditional inference tree analysis to determine statistically significant values of a range of measured variables that predicted average runoff, peak runoff, sediment concentration and sediment production at the microscale. On Chromic Luvisols across a range of vegetation states in semi-arid south-eastern Australia, large changes in runoff and sediment production were related to a hierarchy of different variables and thresholds, but the percentage of bare soil played a primary role in predicting runoff and sediment production in most instances. The identified thresholds match well with previous thresholds found in semi-arid and temperate regions (including the approximate values of 30%, 50% and 70% total ground cover). The analysis presented here identified the critical role of soil surface roughness, particularly where total ground cover is sparse. The analysis also provided evidence that a two-phase mosaic of patches and inter-patches identified via rapid visual assessment could be further delineated into distinct groups of hydrological response, or a multi-phase rather than a two-phase system. The approach used here may aid in assessing scale-dependent responses and address data non-linearity in studies of semi-arid hydrology. Copyright © 2012 John Wiley & Sons, Ltd.