Accurate delineation of watershed boundaries is fundamental for hydrological modeling, water resource management, and Earth system research. However, most global watershed datasets focus primarily on river networks while overlooking the critical role of lakes, which can significantly alter flow paths, watershed boundaries, and hydrological connectivity. This study introduces RiverLakeBasins, a global-scale nested watershed dataset that explicitly integrates lakes into a hierarchical delineation framework by refining and extending the Pfafstetter coding system. Using the MERIT Hydro and HydroLAKES datasets, we simultaneously delineate river-watersheds and lake drainage areas (lake-hillslopes) across multiple spatial levels. We develop reproducible algorithms to optimize stream networks, delineate lake-hillslopes, and reconstruct upstream–downstream flow relationships among rivers, lakes, and lake-hillslopes. Validation against benchmark datasets shows high spatial and topological accuracy: for 97% of shared lakes, the upstream watershed area differs by <5% relative to Lake-TopoCat. On the Qinghai–Tibet Plateau, RiverLakeBasins identifies endorheic lakes with 97.4% accuracy and exorheic lakes with 99.1%, outperforming HydroBASINS. Representative case studies demonstrate improved lake–river connectivity in complex landscapes. By explicitly incorporating lakes and preserving scale-consistent connectivity, RiverLakeBasins provides a global, multi-level dataset to support hydrological modeling, runoff simulation, aquatic ecosystem research, and integrated water resource management.