Updated publication reference for PubMed record(s): 31353208. The translation machinery and the genes it decodes co-evolved to achieve production throughput and accuracy. Nonetheless translation errors are frequent and they affect physiology and protein evolution. Mapping translation errors in proteomes and understanding their causes was hindered by lack of a proteome-wide experimental methodology. We present the first methodology for systematic detection and quantification of errors in entire proteomes. Following proteome mass-spectrometry, we identify in E. coli and yeast peptides whose mass indicates specific amino-acid substitutions. Most substitutions result from codon-anticodon mispairing. Errors occur at sites that evolve rapidly, and that minimally affect energetic stability, indicating selection for high translation fidelity. Ribosome density data show that errors occur at sites where ribosome’s velocity is higher, demonstrating a trade-off between speed and accuracy. Treating bacteria with an aminoglycoside antibiotic or deprivation for specific amino-acids resulted particular patterns of errors. These results reveal a mechanistic and evolutionary basis for translation fidelity.