To understand the uncertainties in a physics simulation code, it is necessary to fully characterize the uncertainties in the physics submodels incorporated in it. I will describe in detail a recent analysis of material-characterization data on tantalum with the goal of determining the parameters of the Preston-Tonks-Wallace model for the plastic deformation of metals. The approach used combines the principles of physics and probabilistic Bayesian analysis with the aim of providing plausible estimates of the uncertainty in predictions made by the resulting PTW model. In the course of this analysis, it is necessary to directly confront systematic errors in each data set. A straightforward approach to dealing with this persistent problem in experimental analysis is presented.