Due to the influence of terrain and other factors, the measured gravity data will be irregularly distributed or missing. Traditional gridding methods, such as kriging interpolation and minimum curvature method, have some limitations in dealing with complex geological structures.Methods: In order to solve the above problems, a gridding method based on diffusion model is proposed in this paper. By simulating the diffusion and denoising process of gravity data, the probability density function of gravity data is learned, so as to generate gravity anomalies that conform to the global statistical distribution.Results: The experimental results of synthetic data show that the diffusion model is effective and accurate in the task of gravity data gridding, and the maximum error and variance of its gridding results are obviously better than those of traditional methods, and it can still maintain a smooth transition in the boundary area, avoiding the obvious oscillation phenomenon of traditional methods in the sparse data area. The experimental results of measured data show that the diffusion model can still accurately restore the detailed structure of gravity anomalies in the case of large missing areas, with small meshing error and strong continuity, which verifies the practicability of this method.Conclusion: Therefore, the gridding method of diffusion model is expected to be applied to the gridding and inversion pretreatment process of measured gravity data in the future, providing more reliable support for geophysical interpretation in complex geological structure areas.