Cerenkov luminescence tomography (CLT) is a powerful imaging technique that allows dynamically and three-dimensionally resolving the metabolic process of radiopharmaceuticals. It uses optical method to detect radiopharmaceuticals with low cost and high sensitivity. However, because of the strong absorption and scatter of biological tissues, the reconstruction of CLT is always converted to an ill-posed linear system which is hard to solve. An accurate and fast reconstruct algorithm becomes a current issue. The traditional reconstruction algorithm based on l2 norm regularization is too smooth and with low accuracy. Some novel sparse reconstruction algorithm has satisfying accuracy and convergence rate, but lose its accuracy for multi-source situation. In this work, a novel CLT method based on forward-backward greedy algorithm is proposed to solve the ill-posed problem. Digital simulations and in vivo experiment were conducted to test the algorithm. The reconstruct results were compared with traditional orthogonal matching pursuit (OMP) algorithm and Tikhonov algorithm. Both the Digital simulations and in vivo experiment show that this approach can reconstruct the distribution of radiopharmaceuticals effectively and accurately.