Tight oil and gas reservoirs have attracted an increasing amount of attentions and have become one of the focus of research field in recent years. Tight sandstones have complex pore structures and narrow pores and throats with pore sizes varying from nanometers to micrometers, and studying flow mechanisms in tight sandstones is of great importance to tight oil/gas reservoir development. Reconstructing digital rock, which can comprehensively represent the petrophysical properties of tight sandstone, is key to simulating the fluid flow in micro/nanopores. This paper proposes a new method of reconstructing 3D digital rock from CT images of tight sandstones based on a deep convolutional generative adversative network (DCGAN), and 3D convolution in the generator and discriminator are adopted to realize reconstruction from 1D data to a 3D digital rock model. The model adopts pore area, volume, spatial distribution and connectivity, Fréchet inception distance score to evaluate the proposed model. Studies show that when the training effect is slightly poor, the generated digital rock model will exhibit noise, which can be reduced by postprocessing; when the training effect is good, DCGAN can accurately reconstruct the 3D digital rock model of tight sandstones, and the reconstructed digital rock is very consistent with the pore size, geometric structure, and connectivity of natural tight sandstones. When multiple 3D tight sandstone CT images are used for training, the DCGAN can learn the pore structure characteristics of entire tight sandstone bodies, which have strong heterogeneous, and the porosity distribution obtained from the generated digital rock is similar to that of the original tight sandstone.