Bioluminescence tomography (BLT) is a promising optical imaging modality, which has the advantages of low cost and high sensitivity. Efficient and stable inverse algorithm is the key to push it into application. To overcome the high ill-posedness of the inverse problem of BLT, we propose a nonconvex L1-2 regularization based on reconstruction method. A convex difference algorithm is used to solve the involved nonconvex functional minimization problem. In each iteration, an alternating direction method of multiplier with adaptive penalty is adopted to solve the problem efficiently. Phantom experiments of single-source and double-source on a digital mouse model are designed to assess the effectiveness and robustness of the proposed method. Comparison study with three typical reconstruction algorithms is also conducted. Simulation results show that the reconstruction results using L1-2 regularization have the optimal location accuracy under different experimental settings.