Since stormwater wash-off of pollutants in urban areas is largely affected by environmental variability, it is very difficult to predict the amount of pollutants transported by stormwater runoff during and after individual rainfall events. We investigated the addition of a random component into an exponential wash-off equation of total suspended solids (TSS) and total nitrogen (TN) to model the variability of runoff pollutant concentrations. The model can be analytically solved to describe the probability distributions of TSS and TN concentrations as a function of increasing runoff depths. TSS data from six Australian catchments and TN data from three of these catchments were used to calibrate the model and evaluate its applicability. Using the results of the model, its potential use to determine the appropriate size of stormwater treatment systems is discussed, stressing how probabilistic considerations should be included in the design of such systems. Specifically, stormwater depths retained by a treatment system should result from a compromise between the recurrence of specific runoff depths and the probability to discharge a target pollutant concentration when such a runoff depth is exceeded.