This seems a rather momentous development-captured on Anthony Watts' site, but originating with Steve McIntyre.
Generally speaking, statements about global temperature that refer to times before about 1900 are somewhat speculative--there are very few actual temperature readings, or long-term readings at one location. So: proxies are used, such as tree rings and other kinds of plant growth.
Tree ring data is at the core of a famous graphic called the "hockey stick"--level cooler temps until the 20th century, then a fairly steep and consistent slope upward in temps in the 20th century itself. This is supposed to show that 1) the 20th century was unusually warm, 2) there was a warming trend during the century, and 3) there was no century of comparable warmth in the last 2000 years (if there was, it probably had natural causes, raising the possibility that any warming today has natural causes). Unless all three propositions are true, it is unlikely that there is any "anthropogenic global warming" to speak of.
Steve McIntyre had earlier demolished a hockey stock based on California trees. Now, after repeatedly trying to get the underlying data, he has apparently done the same thing in the case of some Siberian trees. (Lots of follow-up discussion).
The data set turns out to include a very small number of trees--admittedly a selection of what was available. But how exactly was the selection made--randomly, or what? Researchers have asked repeatedly for complete data sets (which are supposed to be freely available when work is published in refereed journals), often without success. Finally a fairly clear picture is emerging. The small data set was indeed chosen from a much larger possible set--which would not have supported any of the three propositions above. Of course it's possible that some legitimate principle of selection was at work, but then why the secrecy? Why the moves to find "agreeable" data from distant locations, instead of providing the fullest possible picture of nearby locations?