Through the monitoring network for cultivated land quality in county area, distribution and changing trend of the quality should be reflected. Besides, the quality of non-sampled locations should also be estimated with the data of sampling points. Due to the correlation among spatial samples, traditional methods such as simple random sampling, stratified sampling and systematic sampling are inefficient to accomplish the task above. Thus, spatial sampling method based on Kriging estimation variance is presented in this paper. The method considers the spatial structure characteristics of cultivated land and is configurable to restrict the expense or estimation accuracy, making it more flexible to determine the number of samples. What’s more, tessellation polygon is used to densify the network and this helps Kriging interpolation to achieve higher efficiency than traditional methods. Furthermore, inference for the population mean of natural quality indices stays in a very precise level at the same time.