Abstract: From comparisons and connections, the relationships and differences between farmland productivity classes and farmland natural quality grades were explored at county scale in this paper. The farmland productivity grades and farmland natural quality grades in Hengdong County were respectively evaluated by dividing evaluation units with overlay methods, determining quantitative indexes with system cluster analysis, choosing qualitative indicators affecting assessment results with Delphi, computing evaluation factor weights with analytic hierarchy process (AHP) and establishing relational models between indicators and grades with Fuzzy mathematical method. The grade transition distribution was recorded using GIS overlay analysis and the regularities and reasons discussed. The results indicated that farmland productivity grade was one degree lower than farmland natural quality grade in Hengdong County. The proportion of the area of the first five grades of farmland natural quality was 91.29%, but only 74.37% for farmland productivity. The ratio of the area of high, medium and low grades of farmland productivity was 32︰42︰26, but 33︰43︰24 for farmland natural quality. The distribution law of the proportion of area was respectively greater for medium grade field than for high and low grades fields. In terms of percent distribution, significant differences were observed between paddy field and dry land in farmland productivity, higher grades (1, 2 and 3) were mainly distributed in paddy fields. Even distribution among different grades was noted in farmland natural quality. Grade transition percent between farmland productivity grades and farmland natural quality grades clearly showed the extent of differences and spatial distributions of corresponding grades fields. The distribution order of transition percent was plain areas > hilly areas > mountainous areas. Differences in evaluation indicators and weights were the main factors for the differences between two evaluation methods. However, the differences in evaluation objects and sample points also induced some differences in the evaluation results for similar blocks in the two evaluation methods.