A large number of studies in evolutionary biology utilize phylogenetic information obtained from mitochondrial DNA. Researchers place trust in this molecule and expect it generally to be a reliable marker for addressing questions ranging from population genetics to phylogenies among distantly related lineages. Yet, regardless of the phylogenetic method and weighting treatment, individual mitochondrial genes might potentially produce misleading evolutionary inferences and hence might not constitute an adequate representation neither of the entire mitochondrial genome nor of the evolutionary history of the organisms from which they are derived. We investigated the performance of all mitochondrial protein-coding genes to recover two expected phylogenies of tetrapods and mammals. According to these tests, mitochondrial protein-coding genes can be roughly classified into three groups of good (ND4, ND5, ND2, cytb, and COI), medium (COII, COIII, ND1, and ND6), and poor (ATPase 6, ND3, ATPase 8, and ND4L) phylogenetic performers in recovering these expected trees among phylogenetically distant relatives. How general our findings are is unclear. Simple length differences and rate differences between these genes cannot account for their different phylogenetic performance. The phylogenetic performance of these mitochondrial genes might depend on various factors that play a role in determining the probability of discovering the correct phylogeny such as the density of lineage creation events in time, the phylogenetic "depth" of the question, lineage-specific rate heterogeneity, and the completeness of taxa representation.