Over at Bad Science, Ben Goldacre has a nice account of a study which suggests that 'proper' acupuncture might be no more effective than sticking the needles in at random:
The results were fascinating: they set a threshold for “improved” (which was either a 33% improvement in 3 aspects of one score, or 12% improvement on another measuring scale). They found that people having acupuncture were almost twice as likely to reach this level of improvement in back pain as people on medical treatment (which had already let them down for 8 years of course). But even more interestingly, the pretend acupuncture group, where they just bunged needles in any old place with a bit of ceremony, did just as well as the people having proper, posh, theatrical, genuine acupuncture.
A well-designed randomization test can be a most satisfying way of demolishing a theory. By concentrating most of the assumptions into the randomization process they leave very little room for argument.