Multispectral absorption tomography (MAT) is evolving into a mature imaging technique for flow diagnostics due to recent progress in both laser sources and nonlinear tomography. For the absorption-based technique, how to determine and utilize the most informative spectral lines are critical for successful implementation of the technique. In this work, we propose a method to select the optimal combination of spectral lines from the candidate set for MAT. We select the Gram determinant of selected spectral lines in given temperature interval as the cost function of the optimization problem, which can then be maximized by enumerating all the possible combinations of the candidate spectral lines. The numerical studies performed in this work verified the effectiveness of the proposed method, whose purpose was to achieve the best performance for the reconstruction of temperature distribution.