The equivalence between robust worst-case optimisation andvariance-based optimisation has been mathematically derivedunder the assumption that a linearisation is applied and the normsare chosen adequately. Both approaches have been numericallycompared using a simple benchmark problem, i.e. the reduction ofthe size of the PMs in a PMSM while maintaining theelectromotive force. It is found that robust optimisation in thestochastic formulation gives less pessimistic results, since theworst-case might be unlikely to happen. However, thecomputational time is significantly increased whereas theimplementation effort reduced since no (further) derivatives areneeded. The implementation of an affine decomposition facilitatesthe calculation of the derivatives and thus an efficient gradientbased optimisation procedure was obtained. The use of MOR hasbeen shown to be beneficial when a lot of finite elementevaluations are needed, which is the case for Monte Carlo samplingand when using the particle swarm optimisation algorithm.The robustness of the found optima was tested by determiningthe failure rates. It was found the worst-case optimisation