The problems arising in the computed tomography area are well known for their high dimensions and ill-posed. Tikhonov regularization method is developed to reconstruct temperature distribution from deflectometric projection data. Beam deflection angles obtained from moiré deflectograms are integrated to optical path length differences, and a modified regularization method is applied to the linear projection equations. The conjugate gradient method is used to compute the regularized solution for the least-square equations. In numerical simulation, the approach produced reliable reconstructions by computing underdetermined equations and overdetermined equations respectively for asymmetrical distribution. The average errors were 2.1% and the maximum value errors were 5.2% after 10 iterations, which provided a good indication of the precision and convergence of the method. The experimental results of reconstruction from real projection data were satisfactory when compared with the reconstruction results by deflection angle revision reconstruction technique (DARRT) and the direct thermocouple measurements.