This article presents a new approach for [image omitted] or [image omitted] robust model reduction for uncertain systems based on an optimisation algorithm applied directly in the space of reduced-order model parameters. The proposed approach can be applied to both polytopic or affine parameter-dependent models as well as for continuous or discrete-time systems.The developed procedure is capable of computing both fixed reduced-order model as well as uncertain reduced-order model keeping the original uncertainty structure. The resulting reduced-order model is guaranteed to be robust stable.