Estimation of underground contamination is essential to calculate the risks, to make decisions and to evaluate the performance of the clean-up efforts in environmental engineering applications. In this paper, the forecasting accuracy of Backpropagation-type neural networks was tested on various functions and its accuracy was demonstrated. A user friendly self contained Underground Contamination Visualization Package (UCVP) was developed by using the network to establish a 3-D spatial model of the site from the concentration data of test wells without giving any additional information on the soil and contamination source. The package allows the user to find the concentration of contaminants at any depth from the point he selects on the map. In addition, density variation of the contaminant can be inspected at any XY, YZ, and ZX surface by using the three gray-scale plots, which are displayed simultaneously. The operator needs minimal information about modeling of underground contamination and neural networks to visualize the underground plum.