The uncertainty of an electromagnetic device is inherent in its manufacturing process. To consider the various uncertainties, several probabilistic design optimisation techniques, such as robust or reliability-based design optimisation, have been developed. Although a statistical model of uncertainties is extremely important in obtaining an accurate result from a probabilistic design optimisation, most studies on probabilistic design optimisation have assumed these uncertainties to follow normal distributions. However, this assumption may not be valid in several real-world applications. Therefore, this study presents an efficient uncertainty identification method that provides a systematic framework to select the fittest distribution and find its optimal statistical parameters using finite element analysis and experimental data from prototype testing. The Akaike information criterion and maximum likelihood estimation are used for model selection and parameter estimation, respectively. To reduce the computational cost, the kriging surrogate model is used to evaluate the response of the electromagnetic device. The proposed method is applied to a surface-mounted permanent magnet synchronous motor, to identify the uncertainties that produce the additional harmonic components of cogging torque. The results show that this method is a powerful tool in analysing the effect of uncertainties on the performance of an electromagnetic device.