Metamodels are effective for providing fast-running surrogate measures of product or system performance. Instantaneous performance estimation can allow designers to pursue design alternatives interactively, a methodology different from traditional case-based or optimization-based design. Metamodel approximations are generally based on continuous functions, and therefore they can be poor at fitting discontinuous responses. Many engineering models produce functions that are only piecewise continuous, due to changes in modes of behavior or other state variables. This paper shows that a state-selecting metamodel approach can provide an accurate approximation for piecewise continuous response functions. The method is applied to a desk lamp performance model.