It has been widely reported that some genes are expressed at a higher level than others. However, it has not been shown whether each gene is expressed consistently between studies and at the same level compared to all other genes. Here, we examined nine RNA-seq datasets and found that the mRNA level of each gene is indeed consistent relative to all other genes. The mRNA levels were also highly correlated to protein levels measured by different methods, though translational and protein turnover regulation likely prevent a perfect correlation. The consistency of gene expression level implies that there are evolutionary pressures that drive genes to maintain either low or high expression. In order to identify these pressures, we compared expression level to the features of each gene (or associated protein), such as protein function, localization, essentiality and number of interactions. We found many possible pressures; for example, genes involved in translation or the ribosome were highly expressed while genes involved in chromosomal activities, such as replication and transcription, were lowly expressed. Furthermore, we optimized an artificial neural network, using several of these features as inputs, and could predict gene expression level with ~80% accuracy. In conclusion, these results show that gene expression level is likely constrained by several evolutionary pressures, including the biological process and cellular localization of the associated protein.