Abstract: Determining the phase relations between the lung tissue images acquired by means of CT cine scan is a key step to build the 4Dlung model.To solve the problems of existing methods, we propose an improved method for sorting 4D-CT images according to the principle that the adjacent images are most similar.Firstly, the lung tissues are segmented from the initial thorax images using the semiautomatic threshold segmentation algorithm.Then, a benchmark phase is selected and the root mean square error is used as the similarity metric to determine the phase relation between the adjacent couches.Following that, the least square fitting is applied to eliminate the slice spacing error between the adjacent couches.Finally, the lung model reconstruction is carried out.The results indicate that the reconstructed lung model of each phase has a smooth surface with a clear outline, and is consistent with the variation of respiratory movements.Therefore, the proposed method is verified to be effective.Moreover, agood foundation has been laid out for the study of dynamic lung modeling and simulation at the next stage.