Bioluminescence tomography (BLT) is a non-invasive, highly sensitive optical molecular imaging technique, which can reveal the three-dimensional (3D) distribution of the bioluminescent sources inside the tissue through the light signals detected on the surface. The BLT reconstruction problem is ill-posed due to the domination of scattering during the light propagation through the tissue, which results in a challenge to accurately reveal the 3D source distribution. According to the sparse distribution of the bioluminescent sources, the sparse regularization method based on L1 norm has achieved a significant improvement comparing to the traditional L2 norm regularization. Furthermore, due to the spatial aggregation characteristics of the bioluminescent light sources, adopting this feature would further improve the BLT reconstruction accuracy. Comparing to the traditional sparse reconstruction algorithm which takes all unknowns in the solution domain into account, the feasibility of block sparse priori information used for the BLT reconstruction is explored. First, the solution domain is divided into a series of data blocks through analyzing the correlation coefficient between the columns of the system matrix. Then, the block sparse Bayes learning algorithm is used to reconstruct the distribution of the bioluminescent sources. Through the simulation experiment and the mouse in vivo experiment, and compared with those by the traditional sparse reconstruction algorithm based on L1-LS, the results show that the proposed method can effectively alleviate the ill-posedness of the BLT reconstruction problem, suppress noise, and improve the reconstruction accuracy.