In order to address the difficulties in elevation extraction for rugged terrain due to phase undersampling by conventional InSAR system,a phase unwrapping method supported by phase-gradient reconstruction from multi-frequency interferograms is presented.Based on statistical distribution property of wrapped phase,a maximum likelihood estimation technique using multi-frequency phase-gradient is investigated first.Considering the problem of high sensitivity to phase noise under limited number of frequencies in resolving gradient ambiguity,an improved phase-gradient estimation algorithm based on local surface modeling within neighborhood set is presented,and the way to enhance the robustness is further studied from aspects of frequency selection,estimation interval restriction and so on.Finally,experiments using simulated and real DEM data are performed respectively,and the results confirm the validity of the proposed method.