Chronic inflammation is the common pathological phenotype of multiple autoimmune diseases. Characterizing the components of inflammation across diseases is critical to defining and treating diseases based on shared etiology. Recently, single cell RNAseq has made it possible to assay gene expression in thousands of individual cells across diverse conditions. The hope is then to identify common expanded populations across diseases. This goal is hindered by a major analytical challenge: cells from different datasets often group by technical factors, such as platform and read depth, and biological factors, such as source tissue and donor, instead of by cell type. These covariates, though important, confound the identification of shared cell types. We present Harmony, a computational tool to enable common cell type identification across experimental conditions. Harmony projects cells into a shared space in which cells group by cell type rather than dataset specific conditions. Unlike other methods, Harmony can simultaneously account for multiple factors. Moreover, it is the only available algorithm efficient enough to make the integration of ~1 million cells feasible on a personal computer. We use Harmony to power a joint analysis of infiltrating immune cells from inflamed human kidney, synovium, and colon samples. For comparison, we include cells from cord blood and bone marrow, courtesy of the Human Cell Atlas. Through the analysis, we characterize shared and tissue-specific gene expression signatures of immune populations. Harmony is a fast and flexible tool that enables joint analysis of cells across multiple experimental and biological conditions.