Darwinian selection is the driving force of functional evolution. Even at the simplest level of nucleotide sequences, however, the pathways and constraints followed by selection remain very poorly understood. The ongoing avalanche of genome-scale data on sequence divergence and polymorphism is relevant here: combining such data from multiple species with known phylogenetic relationships can help decipher complex evolutionary scenarios. In particular, epistatic selection, i.e. situation when the fitnesses of different alleles at a site are dependent on alleles present at another site, is expected to lead to non-uniform distribution of allele replacements on the phylogeny. I will tell about some of our work in this field. We address a number of questions on genome-level role of selection and epistasis, using genomes and phylogenies of mammals, insects, viruses, etc. Some of the questions asked are: What fraction of amino acid substitutions are positively selected? Do epistatic interactions between substitutions play a major role in evolution of coding and non-coding sequence? Do radical mutations that get fixed in a population lead to a series of subsequent small-effect mutations that alleviate their effect?