Abstract: Objectives: The vertical resolution of crustal structure inverted by using gravity gradients needs to be improved at present. Method: This paper employs wavelet multi-scale analysis method to decompose the vertical gravity gradient data in the Tibetean Plateau, and uses power spectrum analysis method to estimate the field source depth of gravity gradient with different wavelet decomposition orders. Further, according to the average field source depth of gravity gradients of different orders, the gravity gradient signals generated by the Moho of the Tibetan Plateau are extracted to invert the Moho topography. Results: The more detailed structural features at different depths are obtained. Conclusions: The wavelet decomposition results of gravity gradient show that the vertical gradient of disturbance generated by shallow field source less than 10 km depth is scattered in the edge and south-central of the region. The vertical gradient of the disturbance generated by the middle-level field source with a depth of 20km-50km is mainly distributed in the central and southern regions, and the closed circle gradually expands. There are large density bodies in the region of about 60 km depth. In the lower crust and upper mantle below 70 km depth, the density distribution has no obvious characteristics. The inversion results of the Moho show that the terrain of the Moho in the Tibetan Plateau is generally deep in the west and shallow in the east, with an average depth of about 48 km and a maximum depth of about 66 km, located at (78 ° E, 35 ° N). The spatial pattern of the Moho terrain in the Tibetan Plateau is similar to that provided by the CRUST1.0 model, and the correlation (COFF) is about 0.83. Compared with the control point depths, the accuracy of the inverted Moho terrain result is 8.81 km, which is better than 11.4 km of CRUST1.0.