Many outstanding questions about dinoflagellate evolution can potentially be resolved by establishing a robust phylogeny. To do this, we generated a data set of mitochondrial cytochrome b (cob) and mitochondrial cytochrome c oxidase 1 (cox1) from a broad range of dinoflagellates. Maximum likelihood, maximum parsimony, and Bayesian methods were used to infer phylogenies from these genes separately and as a concatenated alignment with and without small subunit (SSU) rDNA sequences. These trees were largely congruent in topology with previously published phylogenies but revealed several unexpected results. Prorocentrum benthic and planktonic species previously placed in different clusters formed a monophyletic group in all trees, suggesting that the Prorocentrales is a monophyletic group. More strikingly, our analyses placed Amphidinium and Heterocapsa as early splits among dinoflagellates that diverged after the emergence of O. marina. This affiliation received strong bootstrap support, but these lineages exhibited relatively long branches. The approximately unbiased (AU-) test was used to assess this result using a three-gene (cob + cox1 + SSU rDNA) DNA data set and the inferred tree. This analysis showed that forcing Amphidinium or Heterocapsa to relatively more derived positions in the phylogeny resulted in significantly lower likelihood scores, consistent with the phylogenies. The position of these lineages needs to be further verified.