Sellafield Ltd is looking to improve the performance of the Magnox Reprocessing plant. To support the improvement programme The National Nuclear Laboratory was asked to use the existing simulation model to understand the impact of individual projects and the combined programme of proposed changes on a performance target. Historically, the approach to identifying a suitable portfolio of improvements has been through experience with the model and then a process of trial and error running the simulation model until the required performance target is met. This is time consuming and limits responsiveness to customer requirements. This paper reports our initial findings on the use of response surface methodology to develop simulation-based metamodel using smaller data sets. The results indicates that these type of metamodels can produce a good estimate for the mean and standard deviation of the performance target.