Background: Wheat landraces and cultivars stored in gene banks worldwide represent a valuable source of genetic diversity for discovering genes critical for agriculture, which is increasingly constrained by climate change and inputs reduction. We assembled and genotyped, using DArTseq technology, a panel of 461 accessions representative of the genetic diversity of Swiss wheat material. The collection was evaluated for powdery mildew resistance under field conditions for two consecutive years and at the seedling stage with 10 different wheat powdery mildew isolates. Results: To identify the genetic basis of mildew resistance in wheat, we developed a k-mer-based GWAS approach using multiple fully-assembled genomes including Triticum aestivum as well as four progenitor genomes. Compared to approaches based on single reference genomes, we unambiguously mapped an additional 25% resistance-associated k-mers. Our approach outperformed SNP-based GWAS in terms of number of loci identified and precision of mapping. In total, we detected 34 (Pm) powdery mildew resistance loci, including seven previously-described and more importantly 27 novel loci active at the seedling stage. Furthermore, we identified a region associated with adult plant resistance, which was not detected with SNP-based approaches. Conclusions: The described non-reference-based approach highlights the potential of integrating multiple wheat reference genomes with k-mer GWAS to harness the untapped genetic diversity present in germplasm collections.