Mitochondria are composed of proteins encoded by both the nuclear and mitochondrial genomes and the coordinated expression of both genomes is essential for energy production. Impaired energy production leads to mitochondrial dysfunction that causes or contributes significantly to a variety of diseases including cardiovascular diseases. To unravel how mitochondrial function fails and to identify therapeutic targets it is necessary (i) to understand how gene expression is regulated between mitochondria and the nucleus and (ii) how this regulation is disrupted in disease. Mammalian mitochondrial ribosomes are unique molecular machines that translate 11 leaderless mRNAs. To date it is not clear how mitoribosomes recognize and initiate translation in the absence of untranslated regions in the mitochondrial mRNAs. Translation initiation in mitochondria shares similarities with prokaryotic systems, such as the formation of a ternary complex of fMet-tRNAMet, mRNA and the 28S subunit, but differs in the requirements for initiation factors. Mitochondria have two initiation factors, MTIF2 that closes the decoding centre and stabilizes the binding of the fMet-tRNAMet to the leaderless mRNAs, and MTIF3 whose role is not clear. We knocked out Mtif3 in mice and show that this protein is essential for embryo development and heart- and skeletal muscle-specific loss of MTIF3 causes dilated cardiomyopathy. We identify increased but uncoordinated mitochondrial protein synthesis in mice lacking MTIF3 that results in loss of specific respiratory complexes by mass spectrometry. Therefore, we show that coordinated assembly of OXPHOS complexes requires stoichiometric levels of nuclear and mitochondrially-encoded protein subunits in vivo. Our ribosome profiling and transcriptomic analyses show that MTIF3 is required for recognition and regulation of translation initiation of mitochondrial mRNAs, but not dissociation of the ribosome subunits. To investigate translation fidelity we created yeast and mouse models with error-prone and hyper-accurate translation, which revealed that translation rate is more important than translational accuracy for cell function and energy production. Our proteomic and metabolomic analyses identified mammalian-specific signalling pathways that respond to changes in the fidelity of protein synthesis and regulate energy metabolism.