A major element in constructing a groundwater model is choosing the parameter values. The traditional approach is to aim for a single best set of values. The parameters used in a model are effective rather than measurable, and this combined with the inherent uncertainties in the modeling process means that there are often many plausible sets of values. A single prediction obtained from a single set of parameter values is not appropriate, but rather the range in predictions from the alternative calibrations should be used. A method is presented for finding the best case and worst case predictions among the plausible parameter sets and is applied to a real case study. Widely different feasible parameter sets were found giving significantly different predictions. ¿ American Geophysical Union 1994