Abstract: In order to overcome the shortcomings of fuzziness, randomness and interaction between the soil corrosion factors of buried pipeline, as well as the low accuracy of prediction with the traditional methods, a prediction model of external corrosion rate was established with 10 influencing factors as the input, and the external corrosion rate as the output based on the field data of corrosion coupons of a buried pipeline. Thereby, the data samples were trained, verified and tested using the Radial Basis Function (RBF) neural network mode, and the key parameters affecting the corrosion were identified through Sobol sensitivity analysis. The results show that the mean square error is 0.000 99 when 10-35-1 type RBF model is iterated to step 2 273, and the correlation coefficients of the training, validation and testing stages are 0.970 7, 0.981 3 and 0.990 1 respectively. Compared with BP, MLR and SVM models, the average relative error of RBF neural network model is 2.07%, indicating that RBF neural network model has some advantages in terms of the external corrosion rate prediction of buried pipeline. The soil resistivity has the maximum effect on the external corrosion rate. Moreover, the soil resistivity, pH value, and Cl^- content significantly interact with other factors, which should be paid much more attention. Generally, the established model can be effectively applied to the external corrosion rate prediction of pipeline, and the results could provide theoretical basis and reference for pipeline integrity management.