Changes in gene expression are thought to underlie many of the phenotypic differences between species. However, large-scale analyses of gene expression evolution were until recently prevented by technological limitations. Here we report the sequencing of polyadenylated RNA from six organs across ten species that represent all major mammalian lineages (placentals, marsupials and monotremes) and birds (the evolutionary outgroup), with the goal of understanding the dynamics of mammalian transcriptome evolution. We show that the rate of gene expression evolution varies among organs, lineages and chromosomes, owing to differences in selective pressures: transcriptome change was slow in nervous tissues and rapid in testes, slower in rodents than in apes and monotremes, and rapid for the X chromosome right after its formation. Although gene expression evolution in mammals was strongly shaped by purifying selection, we identify numerous potentially selectively driven expression switches, which occurred at different rates across lineages and tissues and which probably contributed to the specific organ biology of various mammals. Our transcriptome data provide a valuable resource for functional and evolutionary analyses of mammalian genomes. Overall design: To study mammalian transcriptome evolution at high resolution, we generated RNA-Seq data (~3.2 billion Illumina Genome Analyser IIx reads of 76 base pairs) for the polyadenylated RNA fraction of brain (cerebral cortex or whole brain without cerebellum), cerebellum, heart, kidney, liver and testis (usually from one male and one female per somatic tissue and two males for testis) from nine mammalian species: placental mammals (great apes, including humans; rhesus macaque; mouse), marsupials (gray short-tailed opossum) and monotremes (platypus). Corresponding data (~0.3 billion reads) were generated for a bird (red jungle fowl, a non-domesticated chicken) and used as an evolutionary outgroup. Important note: Fluorophore intensity files were processed with the Ibis base caller (version 1.11). As illustrated in Supplementary Note Table 1 and Supplementary Note Figure 1 of Brawand et al. (Nature 2011), we found that Ibis significantly increased the number of usable reads and drastically reduced the error rate. However, Ibis quality values are not comparable to those from Illumina values, which should be taken into account when using our data and interpreting their quality relative to data established using other base calling approaches.