The guys at RealClimate do a pretty good analysis of Past reconstructions: problems, pitfalls and progress in the context of puncturing some recent contrarian data. They leave out that this exact same argument (and there is a lack of zeal in demanding the same honesty from that side of that debate) applies to plenty of data that has been used suspiciously in numerical models. I will say it again, like I have a dozen times. In any science, if you have 50 million data points and choose 500, that's perfectly valid, but how you choose them and which data points you choose makes a huge difference in the results and far too many climate models have failed the honesty test there. Data, without bias or picking a conclusion and finding data to match, will win here but this has become political because politicians (yes, you Al Gore) started picking data they believed in and declaring that's all there was. Internal boards at competitors will now be atwitter with "denier" talk because they are still in the 19th century regarding actual diversity of thought.