Abstract: An algorithm was proposed for the characteristic of the important wavelengths of the spectral images acquired by electronically tunable filters. The optimal wavelengths are determined by electing the candidates according to the wavelength-specific contributions based on a valid PLSR model firstly, and then followed by an exhaustive analysis of the combinations of the candidate wavelengths. The performance of the algorithm was tested using the spectral images of fresh pork acquired by an acousto-optical tunable filter-based spectral imager and the corresponding chemical indices of total volatile basic nitrogen. The results showed that the proposed algorithm could determine effectively the characteristic wavelengths of the spectral images acquired by electronically tunable filters.