Abstract: Conventional seismic tomography includes ray tomography and wave-equation tomography.As a result of the limitation of theoretical assumptions,the conventional tomographic methods based on raypath are unable to build accurate velocity model in area with complicated geological conditions.Moreover,wave-equation tomography fails to get a good application result,because it needs to establish the initial model with high precision.The method of migration velocity analysis based on Gaussian beam tomography (GBT) depends on the alternately iterative of migration and tomography to achieve velocity update,by which we can obtain the relatively correct velocity model.This method can improve the stability of inversion by constructing less ill-posed Frechet matrix using Gaussian beams’ kernels.The pre-stack Gaussian beam depth migration not only has great advantages in high and steep structure zones,but also can extract angle-domain common imaging gathers (ADCIGs) with high resolution directly.By inputting the ADCIGs to Gaussian beam tomography,we will be able to obtain accurate velocity model.In order to avoid complex artificial pickup in the high-steep structure area,we proposed automatic picking technique to pick up reflection points and the local stratigraphic structural dip.The model test and field data prove that the Gaussian beam tomographic method has high inversion accuracy and stability.