In this study, the use of DoE for training metamodels in simulation-based optimisation of manufacturing systems is evaluated. The evaluation is done through a case study of a real manufacturing system. A simulation model of the system exist and the aim is to train an Artificial Neural Network as a metamodel of the system with as high accuracy as possible. Two training data sets generated using different DoE designs are evaluated and compared to a random training data set. The combination of DoE generated data and randomly sampled data is alsoevaluated.