Ground-water models are embodiments of scientific hypotheses. As such, the models cannot be proven or validated, but only tested and invalidated. However, model testing and the evaluation of predictive errors lead to improved models and a better understanding of the problem at hand. In applying ground-water models to field problems, errors arise from conceptual deficiencies, numerical errors, and inadequate parameter estimation. Case histories of model applications to the Dakota Aquifer, South Dakota, to bedded salts in New Mexico, and to the upper Coachella Valley, California, illustrate that calibration produces a nonunique solution and that validation, per se, is a futile objective. Although models are definitely valuable tools for analyzing ground-water systems, their predictive accuracy is limited. The terms validation and verification are misleading and their use in ground-water science should be abandoned in favor of more meaningful model-assessment descriptors.