The approach of direct inversion of surface wave dispersion data to obtain shallow earth structure becomes popular compared with the traditional two-step surface wave tomographic method. However, some fundamental problems still exist on selecting proper grids to parameterize model. In this study, we apply model parameterization with a multiplegrid scheme to the direct inversion method. A multiple-grid scheme represents a combination of several collocated grids, such as staggered grids, multiscale grids and multiscale-staggered grids. At each iteration, direct inversion is applied to each individual collocated grid to invert for S-wave velocity perturbations that is then interpolated to a predefined base grid using a 3-D B-spline interpolation method. At the end of each iteration, the S-wave velocity perturbations of all collocated grids are averaged together. The averaged perturbations are used to update the 3-D S-wave velocity model that is then used as an initial model for the next iteration. The efficiency of the multiple-grid direct inversion method is demonstrated by applying it to a dense array in Chaohu lake, a special area with the Tanlu fault zone (TFZ) passed through. Synthetic examples confirm a better performance of the multiple-grid direct inversion method than the traditional one. A detailed shallow 3-D S-wave velocity model of the Chaohu lake is obtained after applying the multiple-grid direct inversion method to real data. The velocity model reveals a staggered pattern of velocity variations with low-velocity anomalies in the Hefei basin and eastern Chaohu lake and prominent deep rooted and high-velocity anomalies beneath the TFZ and Yinping mountain, which may associated with the multi-stage tectonic activities in the Tanlu fault. The multiple-grid scheme of this study can further be applied to other tomographic approaches in the near future.