PJMedia points to a decent rundown of the problems that Steve McIntyre had with some of the underlying data in the infamous "hockey stick" graph:

The real damage came when a retired Canadian mining engineer, Steve McIntyre, and a professor at the University of Guelph, Ross McKitrick, started digging into Mann’s methodology, and found flaws in both his statistical analysis and data interpretation, and published a paper describing them in Geophysical Research Letters in 2005. They showed that Mann’s methodology would generate a hockey stick almost independently of the data input, by feeding it spectral noise.

McIntyre eventually got all the underlying raw data and was able to recently replicate some of the work:

The list of 17 sites that was finally sent to McIntyre represented complete vindication. The presence of Yamal and Polar Urals had already been obvious from the Climategate emails, but the list showed that Briffa had also incorporated the Polar Urals update (which, as we saw above, did not have a hockey stick shape, and which Briffa claimed he had not looked at since 1995) and the Khadtya River site, McIntyre’s use of which the RealClimate authors had ridiculed.

Although the chronology itself was not yet available, the list of sites was sufficient for McIntyre to calculate the numbers himself, and the results were breathtaking. Firstly, the URALS regional chronology had vastly more data behind it than the Yamal-only figures presented in Briffa’s paper

But what was worse, the regional chronology did not have a hockey stick shape — the twentieth century uptick that Briffa had got from the handful of trees in the Yamal-only series had completely disappeared.

Direct comparison of the chronology that Briffa chose to publish against the full chronology that he withheld makes the point clear:

It seems clear then that the URALS chronology Briffa prepared to go alongside the others he put together for the 2008 paper gave a message that did not comply with the message that he wanted to convey — one of unprecedented warmth at the end of the twentieth century. In essence the URALS regional chronology was suffering from the divergence problem — the widely noted failure of some tree ring series to pick up the recent warming seen in instrumental temperature records, which led to the infamous ‘hide the decline’ episode.

Remarkably, however, Briffa did allude to the divergence problem in his paper:

These [regional chronologies] show no evidence of a recent breakdown in [the association between tree growth and temperature] as has been found at other high-latitude Northern Hemisphere locations.

The reason for dropping the URALS chronology looks abundantly clear. It would not have supported this message.