Management of groundwater contamination often involves conflicting objectives and substantial uncertainty. A critical source of this uncertainty in groundwater problems often stems from uncertainty in the hydraulic conductivity values for the aquifer. For a remediation solution to be reliable in practice it is important that it is robust over such potential errors. This paper presents the application of a robust multi-objective optimization method on a field-scale pump-and-treat design problem at the Umatilla Chemical Depot site at Hermiston, Oregon. A simple methodology is used to establish plausible realizations of hydraulic conductivity that are then efficiently sampled within the optimization framework using Latin Hypercube sampling. A noisy multi-objective genetic algorithm, developed and tested earlier on a hypothetical aquifer, is then applied to this field-scale case to come up with a set of robust and Pareto-dominant design solutions for the clean up of contaminants (RDX and TNT) in the groundwater. Interactions between the various trade-offs and the inherent uncertainty at the site are analyzed. Finally it is demonstrated that by using such robust multi-objective optimization schemes, it is possible to increase robustness of the optimal solutions without significant increases in costs.