A capacity-based simulation model has been developed to predict components of the field water balance. The model requires daily information such as the atmospheric evaporative demand and rain events, maximum-available soil water storage (MAWS) and root growth characteristics as weil as crop and soil factors. Actual evapotranspiration is modelled using Eagleman's parameterization. The model was tested for upland rice grown on a 3500 m2 field plot and on an undisturbed Iysimeter, both located at Bouaké, Ivory Coast. Good agreement between calculations and observations of actual evapotranspiration (AET) and drainage (D) measured at 15 sites equipped with neutron access tubes and tensiometers demonstrates the reliability of the mode!. This is also independently confirmed against the Iysimeter data. The model was used to study the influence of the soil spatial variability on AET and D at the bottom of the soil profile. This was achieved by considering MAWS as a stochastic process defined by its probability density and covariance functions. Therefore, the model response for the entire field is expressed in terms of probability. It is shown that a deterministic simulation obtained with the mean value of MAWS may lead to unacceptable predictions of the field average AET and D values, the difference with the stochastic mean strongly depending on the degree of variability and on the water supply conditions.