The cultivated land quality grading results have played an important role in the protection of cultivated land, therefore, research on the test method of cultivated land quality grading results has important practical significance. The main purpose of this study was to prove the rationality of the cultivated land quality gradating results by testing cultivated land natural quality index. This paper presented an improved model, and it was established with inverse distance weighting (IDW) matrix which was amended by cultivated land area, then suspected outliers of the improved model and the original IDW model were extacted. In order to determine the accuracy of suspected outliers, a further study was carried out to establish a standard for judging and extracting suspected outliers, which included analyzing the standard deviation of cultivated land natural quality index of cultivated land unit, and selecting a distance which is greater than the threshold distance of biggest Moran index as a buffer radius. This new method was illustrated by using Ningcheng County of Inner Mongolia as a case study, the standard deviation was 78 and the buffer radius was 600 m were obtained. The results indicated that the method using standard to judge suspected outliers can accurately judge determined outliers and exclude most nondeterministic outliers. Besides, the improved model was better than the original IDW model on extracting the determined outliers unit in the same threshold distance, which illustrated that both the area and the distance were important indicators of the quality of cultivated land. This method can provide a reference for testing the cultivated land quality gradating results at county level, and also provide a new way for the application of spatial autocorrelation analysis in related fields.