Spectral curvature (smile effect) is universally existed in dispersive imaging spectrometer. Since most image processing systems considered all spatial pixels having the same wavelength, spectral curvature destroys the response consistence of the radiation energy in spatial dimension, it is necessary to correct the spectral curvature based on the spectral calibration data of the imaging spectrometer. Interpolation is widely used in resampling the measured spectra at the non-offset wavelength, but it is not versatile because the accuracy is different due to the spectral resolution changed. In the paper, we introduce the inverse distance weighted(IDW) method in spectrum resampling. First, calculate the Euclidean distance between the non-offset wavelength and the points near to it, the points number can be two, three, four or five, as many as you define. Then use the Euclidean distance to calculate the weight value of these points. Finally calculate the radiation of non-offset wavelength using the weight value and its corresponding radiation. The results turned out to be effective with the practical data acquired by the instrument, and it has the characteristics of versatility, simplicity, and fast.