Abstract: Due to the uncertainty of the Lagrange empirical parameter, selecting empirical of parameters for diverse observed data sets introduces uncertainty into the results, which weakens the applicability of the inversion method. By using the turning point of the L curve to replace the Lagrange empirical parameter as the regularization parameter, the algorithm focusing on the preconditioned conjugate gradient algorithm has been improved. The underground models have been converted to models with unequally spaced aiming to solve ill conditioned problem as to well as weaken kernel function attenuation. In order to take full advantage of the gravity gradient multiple components, the method of joint five independent measured components of tensor gradient gravity data has been taken with the purpose of meliorating the non uniqueness of inversion results. The effectiveness and reliability of the improved method are validated by the statistical analysis of multiple sets of synthetic models. For the application of the field data, analysis result shows that the improved calculation method is effectively applicable to the inversion of measured gravity gradient data, through inversion of airborne gradiometry data on Australian Kauring test site, we obtained 3D distribution of underground density anomalies. According to the previous results of gravity data inversion, this paper verifies the effectiveness of the algorithm, and discovers more anomaly blocks besides the central anomaly blocks. Our results show that the improved algorithm using field measurements can inverse the distribution of density anomalies, and the inversion results provide more detailed and reliable pattern information for the density anomaly.