Currently, single-cell measurements like scRNA-seq are used to define cell state heterogeneity: for example, to classify T cells into states like Th1, Th2, Th17, Tregs, etc., which in turn may underlie complex immune traits (like autoimmunity). While scRNA-seq has led to a broad range of insights, it lacks the resolution of surface markers, particularly for immune cells, and doesn’t offer a means to precisely sort these populations. But now, recent technological advancements have enabled multimodal measurements; for example, REAP-seq and CITE-seq measure RNA and surface protein expression in single cells. Surface marker and RNA data together may permit more robust, fine-grained definition of cell states. Here, we present one such integration approach: a canonical correlation analysis (CCA)-based method that aligns features along canonical variates—axes of variation shared between the measurements—and projects cells into that space. When applied to REAP-seq data, for example, we can cluster cells in CCA space to identify cell states defined by correlated gene and protein markers. This approach also identifies surface markers to isolate these subsets. We applied this method to a single-cell REAP-seq data set measuring gene and surface protein expression in whole blood cultured with tetanus toxoid and an anti-PD-1 inhibitor. While we could identify broad cell-types (such as monocytes and T cells) with RNA-seq alone, by incorporating protein data with CCA we were able to identify subsets that were not identified when clustering on gene expression alone—e.g., T cells expressing TIGIT or PD-1 surface markers—and their correlated gene expression profiles.