The paper introduces a new computational method called 'Weighted Nearest Neighbor' (WNN) analysis, which integrates multiple data types measured in single cells to define cellular states more accurately. WNN learns the relative utility of each data type (e.g., RNA and protein) for each cell, enabling an integrative analysis of multimodal datasets. The method is applied to a CITE-seq dataset of human white blood cells, constructing a multimodal reference atlas of the immune system. This atlas reveals previously unreported lymphoid subpopulations and improves the ability to resolve cell states. The authors demonstrate how this reference can be used to rapidly map new datasets and interpret immune responses to vaccination and COVID-19. The WNN approach is flexible and can be applied to various multimodal technologies, including paired measurements of RNA and chromatin state. The method is implemented in the open-source R toolkit Seurat and is available for community use.The paper introduces a new computational method called 'Weighted Nearest Neighbor' (WNN) analysis, which integrates multiple data types measured in single cells to define cellular states more accurately. WNN learns the relative utility of each data type (e.g., RNA and protein) for each cell, enabling an integrative analysis of multimodal datasets. The method is applied to a CITE-seq dataset of human white blood cells, constructing a multimodal reference atlas of the immune system. This atlas reveals previously unreported lymphoid subpopulations and improves the ability to resolve cell states. The authors demonstrate how this reference can be used to rapidly map new datasets and interpret immune responses to vaccination and COVID-19. The WNN approach is flexible and can be applied to various multimodal technologies, including paired measurements of RNA and chromatin state. The method is implemented in the open-source R toolkit Seurat and is available for community use.