A reliable pavement performance prediction model is essential for long-term pavement maintenance and rehabilitation planning. There are many factors affecting joint faulting such as heavy traffic, pavement structure, climatic conditions, pavement age, etc. Design features including dowel, base type, pavement thickness, joint spacing, drainage system, shoulder type are also important factors for faulting. So the factors selection has a big effect on the modeling. In this paper, the adaptability of the widely used Back-Propagation Network (BPN) pavement prediction method, using actual joint faulting data is studied. Prediction models with different factors, including 10-factor model, 8-factor model, 6-factor model and 4-factor model, are established and the prediction results are compared. Research outcomes show that the factors that choosing affect the prediction capability and 8-factor model is most effective. Then the proposed factor selection method can effectively support model development.