For reconstructing based on incomplete projection data, an improved max-entropy reconstruction algorithm is presented in this paper. Both least-square and max-entropy are used in this algorithm, and the penalize-function method is used in iterative processing. The numerical simulation of three different tested fields has been done. Compared with conventional three-dimensional reconstruction algorithms, this novel algorithm has a higher reconstruction precision as well as a better astringency, particularly in the case of easy tested field. For the reconstruction of asymmetric complicated tested field, it also has a good result if suitable projective direction and iterative times can be choosen. Finally, the application of this novel algorithm to diagnose the temperature field of arc plasma is presented. (C) 2008 Elsevier GmbH. All rights reserved.