The investigation of the cellular composition and architecture of tissues is key to understanding mechanisms that underlie tissue function and its disruption during disease. Deconvolution is a computational technique used to quantify the cellular composition of complex tissues profiled with transcriptomics technologies. While first-generation deconvolution methods could disentangle only a handful of cell types (mainly human immune cells), second-generation methods can be trained using single-cell transcriptomics data to learn the transcriptional "fingerprints" of any cell type, thereby enabling deconvolution of any tissue, disease context, and organism of interest. Moreover, these approaches can now be applied to spatial transcriptomics data, revealing the architecture of tissues and the spatial distribution of their cellular constituents. However, the intrinsic flexibility of second-generation deconvolution methods poses major challenges to their validation. In this talk, it will be shown how different types of transcriptomic data can be jointly analyzed with deconvolution techniques to chart the cellular organization of complex tissues in health and disease. A major focus will be on methodologies to characterize the tumor microenvironment and especially tumor-infiltrating immune cells, a critical determinant of anti-tumor immunity.