Significant spatial variability of the accumulation of pollutants in soils can make problems in the determination of the borders defining a zone where pollution, according to the applied legal requirements, is excessive. Particular difficulty is caused by a short-distance variability, disturbing the regularity in a spatial distribution of pollution around the source of emission. The paper presents an alternative, compared to traditional interpolation methods, algorithms based on the optimization and the application non-linear neural networks called mixture density network MDN and feature space mapping network FSM.The benefit from the application of this approach is more information referring to the distribution of pollution. This approach allows the estimation of the local variance of the accumulation of pollutants and approximate local distribution. This allows greater extent of taking into account the uncertainty connected with the spatial variability of soil pollution.