As an alternative to a response surface model (RSM) of data from a complicated process, it is sometimes possible to propose a simplified model based upon an understanding of the process physics and chemistry Such mechanistic modeling can yield more insight into process behavior than an RSM The functional form of a mechanistic model is not adjustable, however, in the way an RSM is and the model fit must be tested rigorously, even after statistically significant parameter estimates ape regressed. Because process data are noisy, mechanistic models must be validated with statistical tests. Nonparametric validation methodology is applied here to eliminate inaccurate models and select the mechanistic model that is most accurately descriptive of the etch rate in a plasma etching system.