Sadly not all of the information you might want from a fossil can be obtained just by looking at it. No matter how technological your approach (microscopes, SEM, X-rays, CT scans, synchrotron and the rest) some things, while present, will remain inaccessible. For a start, most vertebrate specimens won’t fit into a scanning electron microscope. CT scanners can only penetrate so much matrix or so much bone. X-rays will only give you a certain level of resolution and so on.
Therefore if you really want to count the LAGs in that femur, or look at those melanosomes in that feather then destructive sampling may be your only option. If you want to do some form of geochemical analysis such as looking at the isotopes then this certainly is your only option. The term might sound rather drastic and imply that you won’t be left with much at the end of it, which while far from the truth, is certainly descriptive. The aim of destructive sampling is to extract the necessary information with the minimum amount of damage, but the crucial point here is that the fossil will be irreparably damaged by this process.
This is typically and understandably seen as a necessary evil in most cases. The damage can be minimised and the information gained can be massive. And of course in many cases it need hardly be noticeable. Isolated dinosaur teeth are for example exceptionally common and in most cases damaged or incomplete, and only the smallest of sample (a few milligrams) is needed to do an oxygen isotope analysis say. It is therefore hand not to justify the sampling of teeth for this purpose.
However, as the specimens become more important and better preserved (and rarer) it becomes harder to justify. It’s no surprise that the first destructive work ever done on Archaeopteryx was only published last year, though it’s still more impressive that several specimens were sampled. Similarly, although there are various feathered dinosaurs, it was hard to justify the destructive sampling necessary to look at the colour of Anchiornis until the method had been well established and it was likely good results would be obtained, and that there were several specimens known so that the loss of information on one does not mean that our only record of this is gone for good.
It might be tempting to argue that anything that possibly damages any fossil (or any scientific specimen) irreparably should not be carried out. After all, technology always increases in scope and accuracy and it is only a matter of time before we could put a whole T.rex under and SEM or put a 5 ton block into synchotron to see what is inside. Science is a steam-roller of a methodology – it takes forever with constant checks, rechecks, corrections, restarts and revisions. A few years or decades will not make so much difference, then will it?
Well, that might be true to a degree, but the obvious counterargument to this approach is that this ensures nothing will ever be done. There will always be another method that’s less invasive, or faster, or cheaper, or provides greater detail coming around the corner and if you wait for one, you’ll wait for the next and the next and the next and no research will ever be performed. In palaeontology we don’t have the luxury of infinite resources or to a degree, such time. We have to get some work done, and if a few specimens have to suffer a little damage to produce a great deal of information, that’s probably no great loss (indeed, on average it’s a gain). So while destructive sampling is hardly the first choice for any specimen, and certainly not ever the choice for some of exceptional historical or scientific importance, it’s a necessary tool in our arsenal and one that is used with care, when appropriate.
Dave Hone's Archosaur Musings 
I think that in most cases, if you can mold/cast the specimen prior to sampling, then a well-designed study using destructive sampling is justified. As a curator who actually has to make these decisions, I look at two main criteria:
1. Is the study well-designed, asking clear research questions that can actually be answered by the proposed destructive sampling (i.e., not exploratory)?
2. Will the destructive sampling yield more new scientific information than it could potentially destroy? In most cases, the answer is yes if the study is well-designed.
If the answers to these two questions are unqualified yeses, then I am usually enthusiastic about the destructive sampling proposal. The only time I have reservations is when we are dealing with a holotype or otherwise “unique” specimen.
Any time a specimen is going to be destroyed, it’d be good to not only have molds/casts, but also CT scans of it. Assuming the whole specimen isn’t full of metallic precipitates, this would provide additional valuable data.
Good point, casting really helps out here!
I also agree with those two primary criteria, but also historically important, exceptionally informative or holotype specimens are deserving of special protection.
Nick – It is true that CT scans would also be helpful, but it is not always possible. For example, isotope sampling of mammal teeth only removes a small amount of material. So I don’t really think its worth the expense of a microCT scan necessarily.
Dave – Thus why I said “the only time I have reservations is when we are dealing with a holotype or otherwise “unique” specimen.” Every specimen is unique in some form, so obviously decisions on destructive sampling are done on a specimen by specimen basis.
Personally, IMHO, teeth are better off being laser scanned for preserving data, but I realize that’s not an option available to everyone. Then again, who cares about mammal chompers? 😉 (j/k)
Well yes of course. I was agreeing with you.
Nick – Agreed, though some mammal studies use microCT to look at enamel thickness, which can vary across taxa.
Dave – sorry, such are the vagaries of written rather than spoken correspondence! Looks like we’re on the same page.
Quite. Especially when I dash off correspondence on a Sunday evening without necessarily reading it through properly! 😉