The research background is the calibration of the magnetic suspension and balance system developed in China. The paper introduces the traditional linear calibration modeling methods, discusses how to overcome its limitation after examining the calibration test data carefully. An improved method of nonlinear fitness to construct the calibration model is proposed. The proposed method uses the rational components of the concept of principal component analysis, and adopts the ideas of cubic spline fitting. Finally, We compare the results of the improved method and traditional methods, and prove the effect of the proposed one.