My mother always said "never trust anything that lacks hard parts." Actually that's a lie - she never said anything of the sort. But she undoubtably would have if she'd read a paper in last week's Nature by Robert Sanson, Sarah Gabbott, and Mark Purnell of the University of Leicester. The title, "Non-random decay of chordate characters causes bias in fossil interpretation" is unlikely to sell many copies of The Sun, but it's the sort of thing that makes paleontologist's hair stand on end.
Generally speaking, paleontologists depend on the organisms they study having biomineralized elements (bones, shells, exoskeletons, etc.) which are preserved through fossilization. Things without hard parts tend not to fossilize, which is unfortunate because these are often the animals that you want to look at most - for example, our distant vertebrate ancestors back in the Cambrian. In some rare instances, however, the circumstances of the animals' death can leave unmineralized soft parts preserved as organic films that record details of their anatomy. The discovery of fossils of this sort from deposits like the Burgess Shale in Canada and the Miaotianshan Shales of China has revolutionized our understanding of the early diversification of the major groups of animals.
As with all fossils, preservation is never perfect; bits tend to decay before preservation can take place, so there are always some characters that are missing. But up til now, paleontologists had always assumed that this process was more or less random; in other words, if a feature was missing in one fossil there was a reasonable chance that it would be preserved in another one. Given enough fossils, you would be able to get a fairly complete picture of the animal's anatomy. This turns out not to be true.
Sanson and his coworkers took specimens of a couple of living analogues for early chordates that will be familiar to anyone who's done a proper biology degree (i.e. one where you actually look at some multi-cellular organisms as well as mucking around with gels and test tubes) - the lancelet (aka amphioxus) and the ammocoete larvae of lampreys - and allowed dead individuals of both to decay under controlled conditions. What they discovered was that decay is not random; bizarrely, the characters that are most useful for determining the evolutionary relationships of these organisms are the ones that tend to decay first and are thus less likely to be preserved. The end result of this is that the fossil may give you the impression that the animal is more primitive than it actually is.
Sanson et al postulate that a lot of the "basal" taxa in phylogentic trees like that of the vertebrates may be nothing of the sort; they are actually more advanced organisms that have undergone "stem-wards slippage," sliding down towards the base of the tree in phylogenetic analyses because of missing data. And that slippage is caused by the differential process of decay. If I worked on fossil invertebrates, I'd be feeling pretty sick by now. But as I work on mammals, which have an abundance of hard bits, I'm off for a hearty lunch.