Three dimensional (3D) deformation can be obtained by using differential interferometric synthetic aperture radar (D-InSAR) technique with the cross-heading tracks data of low earth orbit (LEO) SAR. However, this method has drawbacks of the low temporal sampling rate and the limited area and accuracy for 3D defor- mation retrieval. To address the aforementioned problems, by virtue of a geosynchronous (GEO) SAR platform, this paper firstly demonstrates the expressions of 3D deformation and the corresponding errors in GEO SAR multi-angle processing. An optimal multi-angle data selection method based on minimizing position dilution of precision (PDOP) is proposed to obtain a good 3D deformation retrieval accuracy. Moreover, neural network is utilized for analyzing the accuracy of the retrieved 3D deformation under different orbit configurations and geo-locations. Finally, the proposed methods and the theoretical analysis are verified by simulation experiments. A 3D deformation retrieval accuracy of the order of centimeter-level or even millimeter-level can be obtained by using the selected optimal multi-angle data.