This paper presents an efficient methodology for accurate reliability assessment of electromagnetic devices. To achieve the goal, elaborate surrogated models to approximate constraint functions of interest are generated based on the dynamic Kriging method and a hypercube local window. Then, the Monte Carlo simulation scheme is applied to the surrogate models. This leads to reducing computational cost dramatically without degrading accuracy of the reliability analysis. The validity of the proposed method is tested and examined with a mathematical example and a loudspeaker design.