Most response surface methods typically work on isotropically sampled data to predict a single variable and fitted with the aim of minimizing overall error. This study develops a metamodel for application in preliminary design of aircraft engine nacelles which is fitted to full-factorial data on two of the eight independent variables, and a Latin hypercube sampling on the other six. The specific set of accuracy requirements for the key nacelle aerodynamic performance metrics demand faithful reproduction of parts of the data to allow accurate prediction of gradients of the dependent variable, but permit less accuracy on other parts. The model is used to predict not just the independent variable but also its derivatives, and the Mach number, an independent variable, at which a certain condition is met. A simple Gaussian process model is shown to be unsuitable for this task. The new response surface method meets the requirements by normalizing the input data to exploit self-similarities in the data. It then decomposes the input data to interpolate orthogonal aerodynamic properties of nacelles independently of each other, and uses a set of filters and transformations to focus accuracy on predictions at relevant operating conditions. The new method meets all the requirements and presents a marked improvement over published preliminary nacelle design methods.