Salichos and Rokas point out the inconsistent phylogenetic trees produced by different DNA studies. There is disagreement, for example, on whether gastropods (left, below) are more closely related to bivalves (centre) or scaphopods (right).
While Salichos and Rokas give some answers, part of the problem, in my view, is the common tendency to use maximum-likelihood methods to produce a single phylogenetic tree. The standard algorithms will always produce such a tree of course, but it is important to give the equivalent of error bars, and indicate the range of possible trees supported by a given dataset. Phylogenetic networks, like the one below, are a way of doing this. Occasionally, the data forces us to say “we’re not quite sure” to some questions that have been asked.
Phylogenetic network by Katharina M. Jörger, Isabella Stöger, Yasunori Kano, Hiroshi Fukuda, Thomas Knebelsberger, and Michael Schrödl (see their paper)