This paper introduced the method and principle of a traditional probability neural network (PNN) and an adaptive probability neural network (APNN). Based on inverse problem theory, the question of soil classify is investigated. A new method based on the APNN and RBF neural network is put forward. And an intellectualized analysis system of soil classification is established, consisting of parameter estimation and pat-tern recognition. In the system, the variability of soil physical parameters is thought to be small, whereas variability of mechanics parameters is large. A RBF neural network model is established to reflect mechanics pa-rameters according physics parameters. It can offer a good approach to soil classify by APNN. Examples presented in the paper indicate that this method is neat and effective.