Majority of pixels, in the nature, are non-isothermal in three dimensions, especially for the pixels in meter-scale, tens- meter-scale or hundreds-meter-scale which are paid extensive attention by the researchers in geoscience field. The three-dimensional non-isothermal phenomenon even exists in some pixels in centimeter-scale. For the geosciencific researches, it is significant to determine the component temperatures of a pixel precisely. The airborne WSIS (Wide Spectrum Imaging Spectrometer) data with VNIR (visible-near infrared), SWIR (short-wave infrared) and TIR (thermal infrared) bands were used in the study. First, the components of all the pixels in the image were determined by the linear mixing method. Second, each component emissivity of each pixel was calculated based on an emissivity priori knowledge base. Last, a temperature and emissivity separation algorithm was used to inverse the mean temperature of each pixel, regarded as initial value, the Planck function was linearized to construct a multi-band equation set, and the component temperatures of every pixel were inversed by the Bayesian retrieval technique. The results suggest that the inversion precision of the pixel component temperatures is improved effectively by the Bayesian retrieval technique with the assistance of the VNIR and SWIR hyperspectral remote sensing data.