Abstract: To make up the poor efficiency and low prediction accuracy of traditional prediction methods in the pavement performance prediction problem, this study aimed at the prediction of the pavement condition index (PCI) of the ordinary highway asphalt pavement, and then proposed a PCI prediction model based on random forest algorithm by using the performance indexes, pavement structure, traffic parameters and meteorological data of 1 249 group observations of 9 counties in Beijing. The prediction results were compared with the neural network and support vector machine models. Results show that by comparing and analyzing three quantitative evaluation indexes (root mean square error, mean absolute error and r-square) of different models and visual scatter plots, the robustness and accuracy of the PCI prediction model using random forest algorithm are superior to the neural network and the support vector machine models. It can provide scientific basis for subsequent highway maintenance budget application and decision-making plan formulation, which is of great significance for improving the economic benefits of highway maintenance.