The extent to which natural selection shapes phenotypic variation has long been a matter of debate among those studying organic evolution. We studied the patterns of gene expression polymorphism and divergence in several datasets that ranged from comparisons between two very closely related laboratory strains of mice to comparisons across a considerably longer time scale, such as between humans and chimpanzees, two species of mice, and two species of Drosophila. The results were analyzed and interpreted in view of neutral models of phenotypic evolution. Our analyses used a number of metrics to show that most mRNA levels are evolutionary stable, changing little across the range of taxonomic distances compared. This implies that, overall, widespread stabilizing selection on transcription levels has prevented greater evolutionary changes in mRNA levels. Nevertheless, the range of rates of divergence is large with highly significant differences in the rate and patterns of transcription divergence across functional classes defined on the basis of the gene ontology annotation (primates and mice datasets) or on the basis of the pattern of sex-biased gene expression (Drosophila). Moreover, rates of divergence of sex-biased genes in the contrast between Drosophila species show a distinct pattern from that observed in the contrast between populations of D. melanogaster. Hence, we discuss the time scale of the changes observed and its consequences for the relationship between variation in gene expression within and between species. Finally, we argue that differences in mRNA levels of the magnitudes observed herein could be explained by a remarkably small number of generations of directional selection.