Probabilistic water balance modelling provides a useful framework for investigating the interactions between soil, vegetation, and the atmosphere. It has been used to estimate temporal soil moisture dynamics and ecohydrological responses at a point. This study combines a nonlinear rainfall–runoff theory with probabilistic water balance model to represent varied source area runoff as a function of rainfall depth and a runoff coefficient at hillslope scale. Analytical solutions of the soil-moisture probability density function and average water balance model are then developed. Based on a sensitivity analysis of soil moisture dynamics, we show that when varied source area runoff is incorporated, mean soil moisture is always lower and total runoff higher, compared with the original probabilistic water balance model. The increased runoff from the inclusion of varied source area runoff is mainly because of a reduction in leakage when the index of dryness is less than one and evapotranspiration when the index of dryness is greater than one. Inclusion of varied source area runoff in the model means that the actual evapotranspiration is limited by less available water (i.e. water limit), which is stricter than Budyko’s and Milly’s water limit. Application of the model to a catchment located in Western Australia showed that the method can predict annual value of actual evapotranspiration and streamflow accurately. Copyright © 2012 John Wiley & Sons, Ltd.