The improvement of the robustness of the performance of electrical machines in presence of neglected features, such as manufacturing tolerances and deviations in material properties and parameters, is increasingly demanded. The available optimization procedures do not account for manufacturing tolerances, leaving the decision process incomplete. The paper proposes a methodology for the selection of the more robust design towards manufacturing tolerances among the best candidates, detected by an optimization procedure. Statistical tools for tolerance analysis and worst-case analysis are discussed and adopted for the scope. The proposed methodology is used to select a robust design with respect to torque ripple deviations due to geometric tolerances in the laminations of a Synchronous Reluctance Motor with fluid shaped barriers. The statistical performance analysis and the worst-case analysis provide an estimation of the performance deterioration in the presence of the tolerances in the manufacturing process.