Empirical models are applicable over limited ranges of their predictor variables. The space defined by those ranges, the application domain, is the entire space over which the empirical model is applied. One important assumption is that the model’s predictive behavior is consistent over the entire application domain. This assumption is commonly made for critical heat flux (CHF) models when they are applied in reactor safety analysis. The intention of this work is to demonstrate that the current assessment methods used to justify this assumption may not always identify subregions in the application domain where the model’s predictive capability is degraded. This is accomplished by intentionally placing a nonconservative subregion in a CHF model and demonstrating that the current assessment methods are unable to identify that nonconservative subregion. As the existence of a nonconservative subregion may impact reactor safety analysis, a new method is proposed that does identify the nonconservative subregion. This new method is a multidimensional approach capable of demonstrating if the CHF model’s predictive behavior is likely due to random effects or is due to a degraded predictive capability in a given subregion.