Bioluminescence tomography (BLT) is a promising optical imaging technique that offers an important role in pre-clinical medicine research. However, the core issue of BLT, source reconstruction is still a very challenging ill-posed inverse problem. To overcome the ill-posedness of reconstruciton and obtain accurate quantitative reconstrucitons remains a challenge. For unique and quantitative reconstructions of the internal bioluminescent source, diffusion approximation (DA) and simplified spherical harmonics approximation (SPN) with the Monte Carlo (MC) are compared. The results show that the SP3 model which can balance accuracy and speed is the best model in describing the transmission of photons in the organism. Binding the characteristics which source is distributed in vivo sparsely, a reconstruction using sparse reconstruction by separable approximation (SpaRSA) algorithm is performed for BLT based on SPN forward model. In order to verify the validity of the proposed method, in the digital mouse simulation and real mouse experiment，compared with typical l1_ls algorithm our method has a better performance.