The early diversification of angiosperms is a rapid yet complicated process and thus it renders the phylogenetic analyses of early-diverging angiosperms much difficulty. Plastid and nuclear phylogenomic studies have raised several controversial hypotheses regarding the angiosperm phylogeny, whereas mitochondrial genomes have been largely ignored. In this study, we newly sequenced mitochondrial genomes from 18 angiosperms to fill the sampling gaps in magnoliids, Austrobaileyales, Chloranthales, Ceratophyllales, and early-diverging lineages of eudicots and monocots. A data matrix of 38 mitochondrial genes from 107 taxa was assembled to address this question. Although conflicting phylogenies were recovered in this study from different datasets and analytical methods, congruence was achieved regarding the deep relationships of several major angiosperm lineages: Chloranthales always groups with Ceratophyllales, Austrobaileyales is sister to mesangiosperms, and a previously unplaced clade—Dilleniales—is consistently resolved as a sister to superasterids. Substitutional saturation, GC compositional heterogeneity, and codon-usage bias are suggested as common reasons for the noisy signals that impact phylogenetic reconstructions, and angiosperm mitochondrial genes seem to hardly suffer from these factors. In addition, the 3^rd codon positions of the mitochondrial genes contained more phylogenetic signals than the 1^st and 2^nd codon positions, which might be responsible for the incongruent results recovered among different datasets. Due to the rapid radiation process, the relationships among these early lineages are not well resolved. Nevertheless, this study based on mitochondrial genomes provides additional evidence and alternative hypotheses for the early evolution and diversification of angiosperms.