The invention discloses a bioluminescence tomography reconstruction method based on multi-light-source resolution. Optical property parameters and anatomical structure information are utilized as priori knowledge, and multispectral fluorescent data are combined, so that accuracy of BLT light source reconstruction and resolution of reconstructed images are improved; a self-adaptive feasible region iterative shrinkage strategy is adopted, so that morbidness of BLT reconstruction problem is reduced, and dependence, of reconstruction results, on reconstruction algorithms is lowered. A mixed clustering method combining AP clustering and K-means clustering is provided, so that limitation and defects caused when respective clustering algorithms are adopted solely are overcome. In the method, energy distribution information of grid nodes is used directly, and information such as energy, central coordinates and size of a tetrahedron to which reconstructed nodes belong is fully considered, so that stability of clustering algorithms is improved to greatest extent, and calculation time and complexity of the algorithms are reduced.