Engineering surrounding rock is a highly nonlinear complex dynamic system

and its influence factors are so numerous that single evaluation index can accurately not describe the surrounding rock classification. At present，there are many methods to evaluate surrounding rock classification via considering various indicators together. However，the correlations exist in these evaluation indexes more or less. That is to say，these evaluation indexes are indicators of non-Gaussian distribution; so it is difficult to meet the request that sample layers of probabilistic neural network(PNN) must adopt Gaussian distribution as radial basis function. Given this，we presented a symmetric stable distribution (SaS) which not only has a wider range of mathematical expression than the Gauss distribution，but also it could act as the radial basis function of sample beds in probabilistic neural network (PNN). The model of the surrounding rock classification evaluation in the second-stage project of Guangzhou pumped storage power station can be established based on the SaS distribution. The predicted results show that the model has a good effect and its false-positive rate is 4.55%. This model can provide a new way for predicting the effect of surrounding rock classification in underground engineering.