Ctenophores are a phylum of non-bilaterian marine (mostly planktonic) animals, characterised by several unique synapomorphies (e.g., comb rows, apical organ). Relationships between and within the nine recognised ctenophore orders are far from understood, notably due to a paucity of phylogenetically informative anatomical characters. Previous attempts to address ctenophore phylogeny using molecular data (18S rRNA) led to poorly resolved trees but demonstrated the paraphyly of the order Cydippida. Here we compiled an updated 18S rRNA data set, notably including a few newly sequenced species representing previously unsampled families (Lampeidae, Euryhamphaeidae), and we constructed an additional more rapidly evolving ITS1 + 5.8S rRNA + ITS2 alignment. These data sets were analysed separately and in combination under a probabilistic framework, using different methods (maximum likelihood, Bayesian inference) and models (e.g., doublet model to accommodate secondary structure; data partitioning). An important lesson from our exploration of these datasets is that the fast-evolving internal transcribed spacer (ITS) regions are useful markers for reconstructing high-level relationships within ctenophores. Our results confirm the paraphyly of the order Cydippida (and thus a "cydippid-like" ctenophore common ancestor) and suggest that the family Mertensiidae could be the sister group of all other ctenophores. The family Lampeidae (also part of the former "Cydippida") is probably the sister group of the order Platyctenida (benthic ctenophores). The order Beroida might not be monophyletic, due to the position of Beroe abyssicola outside of a clade grouping the other Beroe species and members of the "Cydippida" family Haeckeliidae. Many relationships (e.g. between Pleurobrachiidae, Beroida, Cestida, Lobata, Thalassocalycida) remain unresolved. Future progress in understanding ctenophore phylogeny will come from the use of additional rapidly evolving markers and improvement of taxonomic sampling.