Holography has been widely used in optical displays, high-security optical encryption, and optical artificial intelligence. Optical multiplexing technologies by utilizing various dimensions of light effectively expand the information capacity and density for holography. In this work, we propose and experimentally demonstrate a novel spatially structured-mode multiplexing holography with the assistance of deep learning algorithms. In the experiment, we utilize Hermite−Gaussian (HG) and Laguerre−Gaussian (LG) modes for example as decoding channels of various holographic images. The results prove that these spatial modes work well as a multiplexing dimension in addition to wavelength, polarization, and orbital angular momentum (OAM) of light. In addition, by designing a specifically computed hologram, multiple spatial modes can be superposed together to compose a single decoding channel, which can significantly enhance the capacity and security for holographic encryption. Our work provides a promising scheme for high-capacity computational holography and information encryption.