ABSTRACT The design of groundwater contamination remediation based on hydraulic head gradient control method determines the locations of the pumping wells and their pumping rates. In a heterogeneous medium such a design will be sensitive to the spatial characteristics of the underlying geological parameters. The geological uncertainty is due to the heterogeneity of the hydraulic conductivity of the porous medium. Under conditions of uncertainty, incorporation of transmissivity fields with spatial characteristics that most influence the design will reduce the sensitivity of the design. A new class of artificial intelligence technique known as neural networks has been identified as appropriate for pattern association tasks. A neural network based screening tool is being developed to identify transmissivity fields with such spatial characteristics. The ongoing research embraces training a neural network to learn the association between transmissivity fields and their impact on the design, and using the trained network to classify randomly generated feasible transmissivity fields according to their level of impact on the design.