Abstract: Previous quantitative methods for classifying expansive soil such as “Fuzzy” and “Grey Cluster Analysis” have certain shortcomings．Both of them rely on artificial factors and thus may influent the true results of the soil classification．A new method of Artificia1 Neura1 Network (ANN ) model presented in the paper can well solve the difficulties and it has good objectivity and excellent disturbance resistance．The model is applied for classifying soil samples of D section in Ning-Lian high-grade highway．The results are well in accord with those by “Fuzzy” and Grey Cluster Analysis” methods．The ANN method has been approved available and reasonable in practice and may become an effective way on determining expansive soil’s properties quantitatively．