Abstract: Because the damping least square method is only suitable for a simple model, when the media has multiple layers, inversion is effected by multiple solutions, sometimes even no convergence occurs, and the inversion is very time-consuming. For this purpose, regularization idea is used to introduce the model-constrained inversion, in the cause of iteration, the adaptive regular factor of the regularized inversion algorithm is calculated adaptively based on the relation of the data objective function and the model objective function; it can result in a stable convergence in the iteration course of the inversion. In this paper, the inversion used quasi-Newton method to update Jacobian matrix, it reduced greatly the time required for the inversion, and the typical spreadsheet three and multi-theoretical model as an example to inversion is proved less demanding on the initial model of the adaptive quasi-Newton regularized inversion algorithm. The inversion result also show that the adaptive regularized inversion is steady and reliable and has good fitting effect, fast convergence and strong adaptability.