Parameterization of a hydrogeologic flow and transport model is a principal factor governing the accuracy of predictions of future system behavior. Precision (minimal uncertainty or error variance), in turn, typically depends on an ability to infer the values of distributed system properties from historical measurements of system state through the model calibration process. When such data are scarce, or when their information content with respect to parameters that are most salient to predictions of interest is weak, the uncertainty associated with these predictions may be high, even if the model is " calibrated" . Current modeling practice must recognize this condition and suggest a path toward improved predictive accuracy by identifying sources of predictive uncertainty and by determining what observation types will most reduce this uncertainty. The present paper illustrates a combination of methods that can be used for this purpose, all of which are readily implemented as an adjunct to calibration of highly parameterized models. Both linear and nonlinear methods of analysis are discussed. The former furnish a variety of statistics pertinent to predictive uncertainty levels, including specific and unique contributions made by different parameter types, and the worth of different observation types toward reducing this uncertainty. The latter accurately characterize the uncertainty of critical model predictions while helping identify mechanisms for how relatively unlikely (though possible) system behavior can still occur. The methods can be applied to the prediction of specific discharge made for a saturated zone flow model. The model predictive uncertainty, and by inference the uncertainty of other predictions related to site scale flow and transport, is thereby assessed.