Tumors are complex ecosystems comprising diverse cell types actively participating to carcinogenesis, tumor progression, and treatment response. Understanding the tumor microenvironment (TME) dynamics has become of primary importance, especially with the increasing clinical implementation of immunotherapy. Low and high-throughput single cell and spatial technologies are providing high-resolution strategies for the study of the tumor ecosystem. However, their cost and complexity limit widespread use. Bulk transcriptomics has become a widely used strategy to obtain the expression profile of large cohorts of tumors or preclinical models. Several methods implementing a deconvolution analysis have been developed to estimate from bulk transcriptomics the prevalence of multiple cell types to reconstruct the tumor ecosystem composition. In this chapter, we introduce deconvolution analysis, the main steps, the recent advancements, and open challenges. Our emphasis lies on robust benchmarking methodologies, highlighting the importance of clear parameter definition and appropriate metric selection for reliable results across different software tools. Using CIBERSORTx and BayesPrism, we conduct a practical analysis on triple-negative breast cancer (TNBC) datasets from The Cancer Genome Atlas (TCGA) dataset. We illustrate the impact of various factors such as preprocessing methods, reference datasets, and software choice on deconvolution outcomes. Integrating insights from benchmarking analyses and real-world applications, we provide guidance to optimize and control for the quality of deconvolution analysis, weighting both its potential and limitations. Deconvolution analysis can contribute to unravelling the complexities of the tumor microenvironment, but further research is needed to enhance accuracy and reproducibility.