According to Code for highway engineering geological investigation (JTG C20—2011), three indices, namely, free expansion rate, plasticity index, and water content of soil under standard moisture absorption, were selected as the factors for synthetic evaluation of expansive soil. The swelling potential of expansive soil was divided into four grades: non-expansive, weakly expansive soil, moderately expansive, and strongly expansive. With the soil samples along a highway as an example, the ordinal Logistic regression model of swelling potential classification was established by using SPSS software and was applied to testing other cases. The predicted results were in good agreement with the actual. The results indicate that the ordinal Logistic regression model performs excellently, and can objectively reflect the complicated situation of expansive soil classification, thus is of good prospect in practical engineering.