The study addresses the issue of decoy bias in molecular docking, which can significantly affect the evaluation of docking performance. To address this, the authors have developed a directory of useful decoys (DUD), which consists of 2,950 ligands for 40 different targets, each paired with 36 decoy molecules that are physically similar but chemically distinct. This results in a total of 98,266 compounds. The DUD database was created to ensure that decoys closely match the physical properties of the ligands while being topologically dissimilar, thereby reducing bias in enrichment calculations. The authors compared the performance of DUD with other decoy sets, such as the MDDR database and those used by Rognan and Jain. They found that DUD provided significantly better enrichment factors compared to these smaller, uncorrected databases, indicating that the latter sets were biased due to differences in physical properties between ligands and decoys. The study also included a forty-by-forty cross-docking analysis, where the enrichments of each ligand set were compared across all 40 targets, providing a specificity metric for docking screens. The results suggest that DUD is a more stringent benchmarking set for evaluating virtual screening performance. The study highlights the importance of matching the physical properties of decoys to ligands to avoid bias and provides a valuable resource for researchers in the field. The DUD database is freely available online at http://blaster.docking.org/dud/.The study addresses the issue of decoy bias in molecular docking, which can significantly affect the evaluation of docking performance. To address this, the authors have developed a directory of useful decoys (DUD), which consists of 2,950 ligands for 40 different targets, each paired with 36 decoy molecules that are physically similar but chemically distinct. This results in a total of 98,266 compounds. The DUD database was created to ensure that decoys closely match the physical properties of the ligands while being topologically dissimilar, thereby reducing bias in enrichment calculations. The authors compared the performance of DUD with other decoy sets, such as the MDDR database and those used by Rognan and Jain. They found that DUD provided significantly better enrichment factors compared to these smaller, uncorrected databases, indicating that the latter sets were biased due to differences in physical properties between ligands and decoys. The study also included a forty-by-forty cross-docking analysis, where the enrichments of each ligand set were compared across all 40 targets, providing a specificity metric for docking screens. The results suggest that DUD is a more stringent benchmarking set for evaluating virtual screening performance. The study highlights the importance of matching the physical properties of decoys to ligands to avoid bias and provides a valuable resource for researchers in the field. The DUD database is freely available online at http://blaster.docking.org/dud/.