In theory, the algorithm for detecting dark matter is straightforward: 1) wait until a dark matter particle hits a nucleus in your detector hard enough to produce a visible recoil, 2) count the events and collect the Nobel prize, or set a limit on the dark matter scattering cross section on nucleons. The reality is more complex. Typical models of dark matter models predict the largest signal near the energy threshold of the detector where it is susceptible to all kinds of spooky background and noise. For this reason, the field of dark matter detection, with multiple contradictory claims and a good deal of bad blood, reminds of medieval England at the time of the Wars of the Roses. The latest claim of a dark matter signal from the CDMS experiment brings a new hope but also adds to the confusion.
For most of the previous decade CDMS was the most sensitive direct detection experiment. Their primary target was germanium, but they also had a number of silicon target detectors. The latter are more advantageous to study light dark matter - with mass of order 10 GeV - because silicon nuclei are lighter than the germanium ones, and thus are more prone to get a kick from a light dark matter particle. In the analysis of 56 kg*days of data collected in 2006-2007, posted on arXiv just yesterday, no event passes the cuts designed to separate dark-matter-like recoils from background. To everyone's surprise, CDMS just announced that in the 124 kg*day of data collected in 2007-2008 three events the cuts, while the expected background is 0.4 event. In the past, several underground experiments (DAMA, CoGeNT, CRESST) have detected an excess of events, but it's the first time an excess of this magnitude appears in a low-background apparatus. The probability for the background to produce 3 events is 5%, which would amount to a 2 sigma fluctuation. On the other hand, testing the background hypothesis against the one of light dark matter with the mass of 9 GeV and the cross section of 2*10^-41cm^2 prefers the latter at the 3 sigma level, apparently because the recoil energies and the ionization yields of the events perfectly fit the dark matter hypothesis. Statistics is not an exact science, so you can take either of these two numbers as an estimate of the significance of the CDMS signal, depending on your priors and your allegiance. The significance is probably smaller than 3 sigma anyway if the latest data are combined with the previous silicon and germanium data, which show no signal in the low mass range.
However, the precise significance is not the most important issue here; in the end, we sometimes shrug off 9 sigma signals. To the right, what looks like Pollock's painting is in fact a summary of best-fit signal regions and limits from various underground experiments in the dark matter mass vs. cross section parameter space. The most worrying aspect of the CDMS result is that the signal region seems comfortably excluded by the limits from Xenon-10 and Xenon-100 experiments (the green lines in the plot). To reconcile these results one must either assume a serious systematic issue with the xenon analyses, or consider more exotic dark matter models, for example the xenophobic ones where the effective coupling to xenon nuclei is suppressed. On the other hand, the region of the parameter space preferred by CDMS is consistent with the earlier
detection claim by the germanium target detector CoGeNT.
So, dark matter, a fluke, or a fundon? Unfortunately, the past experience with direct detection experiments suggests that we will not learn the definitive answer anytime soon.