We propose a percentile half threshold pursuit algorithm (PHTPA) for applying optical molecular imaging modality to bioluminescence tomography (BLT). The BLT reconstruction problem is modeled as an L_1/2 regularization problem that can be solved by combining the subspace pursuit (SP) and percentile threshold methods based on the iterative half threshold algorithm (HTA). Several simulations are run on the digital mouse model to evaluate the validity and astringency of PHTPA. The simulation results demonstrate that PHTPA produces more accurate reconstruction results in different source settings when compared with the original HTA and iterative reweighted algorithms.