Gravity data embody the superimposed effect of all the bodies with uneven densities from the earth surface to the deep subsurface. And gravity data with different scales are needed for different tasks. The two-dimensional empirical mode decomposition can be adapted to nonlinear or non-stationary signals to achieve the multi-scale decomposition. In this study, the authors applied the two-dimensional empirical mode decomposition to the multi-scale analysis of gravity data. With this method, the gravity data can be decomposed into different intrinsic mode functions and one remaining component. Then, the radial logarithmic power spectrum of various components are computed to get the approximate depth of various sources. Theoretical density model data and real data test support the technical feasibility of the method.