Bioluminescence　tomography　(BLT)　is　an　emerging　imaging　technique　for　earlier　disease　detection,　drug　discovery,　etc.　However,　BLT　reconstruction　is　a　severely　ill-posed　problem.　Since　multispectral　information　provides　the　depth　information　of　bioluminescent　source　and　improves　the　reconstruction　quality,　therefore,　multispectral　BLT　has　attracted　much　attention.　In　the　paper,　a　Huber-Markov　random-field　regularization　for　multispectral　BLT　is　incorporated　into　a　Bayesian　framework　to　reduce　the　ill-posedness　of　BLT　and　localize　the　bioluminescent　source　accurately.　For　practical　application,　the　optimal　permissible　source　region　was　used　to　reduce　the　ill-posedness　and　improve　the　computational　efficiency.　Finally,　the　performance　of　the　proposed　algorithm　was　validated　by　a　heterogeneous　3-D　micro-CT　atlas　and　a　mouse-shaped　phantom.　Reconstructed　results　demonstrate　feasibility　of　the　proposed　algorithm.　Copyright　©　2010　ACM.