The uncertainty of electric machines and drives is inherent in its manufacturing and assembly. Robust design optimization finds the design under these uncertainties that results in a minimal variance while satisfying all design constraints. To obtain an accurate result, the statistical model is significantly important. Moreover, several uncertainties from a single source can be involved owing to multiple components in the electrical machines and drives. However, this can drastically increase the numerical cost for the conventional robust design optimization technique. In this work an uncertainty characterization method is developed by using a percentile bootstrap interval to consider the experimental results from few prototypes and a kriging surrogate model to reduce the computational cost. Then the sample-based robust design optimization is applied to the surface-mounted permanent magnet synchronous motor. It is seen that the developed methods can efficiently work to minimize both the mean and variance of the cogging torque while satisfying other design constraints.