Popular knowledge of the population structure and genetic diversity of a plant species is essential for designing improvement strategies. The genotyping-by-sequencing (GBS) approach has been used to simplify complex genomes and has become a popular high-throughput molecular tool for selecting and breeding many crop plants, including those with large genomes. This study aimed to construct a core collection of walnut (Juglans regia) germplasm using the GBS approach. A diversity panel of 87 walnut initial genotypes, including 25 landraces, 12 cultivars, and 50 seedling populations, mostly native to the Gansu Province of China, was subjected to GBS. A total of 110,497 high-quality SNPs were identified and used for determining distinct clusters and an optimum number of subpopulations. Structure analysis divided the genotypes into three distinct groups, which coincided with their collection site and year, suggesting a certain degree of separation in the geographical origin and pedigree among the three groups. To maximize germplasm utilization, the genotypes were posteriorly grouped according to the subgroups obtained through GBS analysis. To minimize subsample redundancy, the core collection was designed using a set of 6540 SNPs distributed across all 16 chromosomes. Finally, a core collection comprising nine walnut genotypes (10% of the entire genotype set), including five cultivars, three seedling populations, and one landrace, was assembled. Genetic structure analysis indicated that the core collection has an uneven distribution in the landrace collection, which could be related to environmental conditions, and the genotypes of the landrace collection are similar. Overall, the results of this study and the establishment of the core collection will facilitate the improvement of walnut in future breeding programs.