Abstract: Cancer genomes contain a plethora of information about the complex processes of mutation and selection that formed them. Genomic changes conferring a selective advantage to developing cancer cells drive successive waves of clonal expansion, shaping new cancer clones or subclones. While the existence of genetic heterogeneity within cancers is recognized, systematic insight into the subclonal architecture of cancer is currently lacking. We developed multiple bioinformatic algorithms to characterize the subclonal architecture of cancers from their whole-genome sequences, and we apply these to 21 breast cancers. As a first case-study, we aimed to characterize subclones and their phylogenetic relationships in one ER+ breast cancer sequenced to 188-fold sequencing depth. This tumor’s genome is hypodiploid with relatively few copy number changes and shows a mutator phenotype with 70,692 mutations genome-wide. We developed an algorithm to detect large-scale subclonal copy number changes in tumor genomes, using allele frequencies of germline heterozygous SNP loci. Leveraging information of the 1000 Genomes Project to determine allele ratios across haplotype blocks, we can detect rare subclonal copy number changes and we can accurately quantify the fraction of cells in each subclone. In the deep-sequenced tumor, a deletion of chromosome 13 in 69 % of tumor cells was detected, as well as evidence for a tetraploid subclone (14 % of tumor cells) with single-copy deletions on 9 different chromosomes and deletion of two copies of chromosomes 2 (one from each parent) and chromosome 7 (both from the same parent). These analyses did not require massive sequencing depth, as this could all be detected as well when only 16 % of the sequencing data was used. We modeled the patterns of clonal and subclonal single nucleotide mutations using a hierarchical Bayesian Dirichlet process. This model performs well on simulated data and simultaneously identifies the number of subclones, the fraction of tumor cells and the contributing proportion of somatic mutations. In the example above, 27000 mutations were clonal, 16000 mutations were found in 65 % of tumor cells, and 11000 mutations were found in 18 % of tumor cells. Finally, using novel methods to phase mutations relative to each other and to heterozygous SNP loci, we are able to discern the phylogenetic relationships between the above mentioned subclones. These analyses indicate a complex picture of ongoing clonal expansions of multiple competing subclones that are each shaped by distinct genomic changes. Applying the methods above to 20 breast cancer genomes sequenced to 30-40 fold coverage depth, we gain insight into the subclonal landscape of breast cancer. In 19 of these 20 genomes, we found conclusive evidence of subclonal mutations and/or copy-number changes and many of them show a complex subclonal landscape. This study paints a dynamic picture of on-going evolution and competition between subclones in breast carcinoma.