The traditional soil classification ignores gradual changes of intrinsic properties and continuous spatial changes of soil properties. The fuzzy k-mean (FKM) has been applied in soil classification by McBratney and de Gruijter (1992). In this paper, the results of the FKM are compared to soil groups classified by using Unified Soil Classification System (USCS) for investigating the engineering properties of soil. Additionally, the fuzzy similarity of grain-size distribution (FSGSD) is presented to improve the ability of FKM in soil classification. The validity of classification with FKM can be up to 87% and with FSGSD up to 93%. Since one-level FKM cannot identify the different features of soil groups significantly, the nested FKM and nested FSGSD method is also presented to improve the ability of classification in sub-groups. The nested FKM can identify the soil groups when the group number is over four. The comparison on results of the nested FKM and nested FSGSD is also performed, and it is found that the nested FSGSD remain to be superior to nested FKM.