Crop reference genomes have greatly advanced gene mapping and marker-assisted breeding. However, a single reference genome cannot fully capture the genetic diversity within a crop species. Pangenomes—comprehensive catalogs of genetic elements within a species or genus—are transforming crop genomics by enabling the accurate detection of complex yet important genomic variants such as large structural variants (SVs). Watermelon (Citrullus lanatus subsp. vulgaris) is among the most economically and nutritionally important fruit crops worldwide. Despite considerable phenotypic diversification, watermelon has a narrow genetic base due to long-term selection for fruit-quality traits. In contrast, its wild relatives harbor abundant genetic diversity, particularly resistance genes that are often diverged or even absent in cultivated accessions. Therefore, a pangenomic framework is essential to comprehensively explore watermelon genetic diversity and advance molecular breeding.To address this, I first generated reference genomes for three wild species widely used in breeding and re-sequenced representative cultivated and wild accessions to construct a gene-based super-pangenome. This resource revealed extensive gene presence-absence variation, including resistance-related genes enriched in wild relatives, provided strong genomic evidence supporting Kordofan melon as the direct progenitor of dessert watermelon, and yielded insights into the domestication of flesh bitterness, sweetness, and coloration traits. I then expanded the resource by constructing a graph-based pangenome from 135 newly assembled near-gapless genomes spanning all seven extant Citrullus species. This graph pangenome provided a more complete catalog of the Citrullus gene repertoire and enabled accurate genotyping of nearly one million SVs across ~900 cultivated and wild accessions. This comprehensive variation map uncovered SVs under selection during watermelon domestication and improvement, improved trait-mapping resolution, and identified a causal 1,258-bp copy number variant in the ClFCI1 promoter that regulates flesh color intensity in a dosage-dependent manner. Additionally, I demonstrated the utility of the graph-based pangenome for genomic selection, consistently achieving high prediction accuracy for a range of fruit-quality and disease-resistance traits. Overall, this work establishes a population-level super-pangenome for watermelon, demonstrates the value of graph-based pangenomes as next-generation references, and delivers transferable markers and models for precision breeding, offering a blueprint for future crop pangenomics.