Dataset supporting the manuscript "Three-Dimensional Digital Rock Reconstruction with Multi-Mineral Components Using an Improved Wasserstein Generative Adversarial Network with Gradient Penalty" (submitted to Water Resources Research).Contents: (1) 256 segmented digital rock slices (TIFF) derived from Qemscan multi-mineral imaging of a carbonate reservoir rock in China, used to train the improved WGAN-GP model; (2) 33 three-dimensional multi-mineral digital rocks generated by the trained WGAN-GP model (NumPy .npy volumes), underlying Figures 5-7 and Table 2 of the manuscript. The multi-mineral system includes sodium feldspar, quartz, calcite, illite, chlorite, pyrite, rutile, dolomite, anhydrite, other minerals, and pore space. See README.md for details.