Integral imaging, as an excellent light-field three-dimensional (3D) imaging technique, is considered as one of the most important technologies for 3D encryption because of its obvious advantages of high robustness, security, and computational feasibility. However, to date, there is no effective cryptanalysis technology for the light-field 3D cryptosystem. In this Letter, a cryptanalysis algorithm based on deep learning for the light-field 3D cryptosystem is presented. The 3D image can be optically retrieved by the trained network model without encryption keys. The experimental results verify the feasibility and effectiveness of our proposed method.