Abstract: In order to explore the optimal grid scale and sampling design optimization method for soil heavy metal pollution investigation, soil samples were collected and analyzed in farmland near an old industrial park in central and southern China. Four meshes of different scales, including 70 m×70 m, 100 m×100 m, 160 m×160 m and 200 m×200 m, were constructed, and the estimation accuracy of Cd polluted spatial location in soil was analyzed and verified by estimation test of Matheron moment estimation and geostatistical method. The results indicated that the estimation accuracy of polluted spatial location was similar for 70 m×70 m and 100 m×100 m grid, and higher than 160 m×160 m and 200 m×200 m grid. By encrypting the sampling points on the uncertain region of 100 m×100 m grid, the estimation accuracy of polluted spatial location increased from 78.89% to 86.25%, and the number of investigation samples decreased by 35% for 70 m×70 m grid, not only reducing the cost of sampling and testing, but also improving estimation accuracy of polluted spatial location effectively.