Ancestor-descendent relations play a cardinal role in evolutionary theory. Those relations are determined by rooting phylogenetic trees. Existing rooting methods are hampered by evolutionary rate heterogeneity among lineages or the unavailability of auxiliary phylogenetic information. In this thesis I propose two novel rooting approaches, each approach applicable to address different research questions. In section 4, I introduce a general method to infer the roots of phylogenetic trees, without assuming prior knowledge about phylogenetic relations among the studied lineages. The method, named Minimal Ancestor Deviation (MAD), takes as input any type of unrooted tree and infers the most likely root using branch length and topological information contained in the tree. When applied to biological datasets, I show that MAD is more accurate and more robust to known confounding factors than existing methods. In the next sections, I use Ancestor Deviations (r) in a phylogenomic context to infer the roots of species trees, using whole genomes for the inferences. The approach is grounded in a statistical framework that evaluates all candidate roots of the underlying species tree and formally tests the relative strength of competing root hypotheses. This phylogenomic rooting approach uses information from multiple gene trees and does not require knowledge of the species tree, making it suitable for root inferences even in face of reticulated evolution. When applied to biological datasets, our approaches reveal evidence for: 1) the origin of photosynthesis in the ocean; 2) the anaerobic and chemolithoautotrophic lifestyle of the last common ancestor of proteobacteria; and 3) the chimeric nature of modern archaea genomes.