PURPOSES: The objective of this study was to develop an asphalt pavement response model for a subsurface cavity section using the 3D finite element method and a statistical approach. METHODS: It is necessary to analyze the structural behavior of asphalt pavement with a subsurface cavity to evaluate the degree of risk for a road cave-in. A 3D finite element model was developed to simulate the subsurface cavity underneath asphalt pavement and was verified using the ILLIPAVE program. Finite element analysis was conducted for asphalt pavement sections with different asphalt layer thickness/modulus, and cavity depth and length, to generate the artificial pavement response database. The critical pavement response considered in this study was the tensile strain at the bottom of the asphalt layer because fatigue cracking is the main cause of road cave-in. The relationship between the critical pavement response and influencing factors was investigated using the pavement response database. The statistical regression approach was adopted to develop the asphalt pavement response model for predicting the critical pavement response of asphalt pavement with a subsurface cavity. RESULTS : It was found from the sensitivity analysis that the asphalt layer thickness or modulus, and cavity depth or length, are the major factors affecting road cave-in incidents involving asphalt pavement. The asphalt pavement response model showed high accuracy in predicting the tensile strain at the bottom of asphalt layer. It was found from the verification study that the R square value between finite element model and pavement response model were 0.969 and 0.978 in the cavity and intact sections, respectively. CONCLUSIONS: The work reported in this paper was intended to figure out the pavement structural behavior and to develop a pavement response model for the occurrence of cavities underneath asphalt pavement using 3D finite element analysis. In the future, critical pavement response will be utilized to establish the criteria of risk of road cave-in based on various different conditions.