In order to achieve efficient, rapid and accurate evaluation of slope stability, multiple evaluation indicators that affect slope stability are considered. There is a certain degree of multivariate collinearity among the indicator variables, which may cause overlap of data and information, leading to certain errors in slope stability evaluation. In this study, a principal component analysis (PCA) was introduced to decorrelate and reduce the dimensions of the slope stability-related variable data, and three comprehensive indicators were extracted to comprehensively evaluate the slope stability. After the principal component analysis, the indicators were mutually independent, can better meet the Gaussian distribution requirements in the Radial Basis Neural Network (RBFNN). On the basis of principal component analysis, the slope stability evaluation RBFNN model is established and applied to 32 sets of typical slope measured data in China. The simulation results show that the principal component analysis-RBFNN model is wrong in 6 different learning situations. The judgment rates were 6.25%, 6.25%, 6.25%, 9.38%, 15.62%, and 15.62%, respectively. At the same time, the evaluation results were discussed and tested. It shows that the principal component analysis-RBFNN model can provide a new idea for slope stability evaluation. [ABSTRACT FROM AUTHOR]