I was listening to NPR and there was a report on Hal Pashler, Chris Baker, and Ed Vul's criticisms on the use of fMRI. The report is essentially a recap of the Vul et al paper that was been debated in neuroimaging circles for almost a year. The original version of the paper, omniously named Voodo Correlations in Social Neuroscience, and can now be read in it's final, less incituous form here.
Essentially the paper argues that the results of social neuroscience papers using fMRI techniques are simply too good to be true. The NPR report unfortunately generalizes this to all fMRI studies although the criticism was initially only leveled at the social neurosciences. Setting aside huge problems in data collection and the ethics involved in publication, the Vul et al paper (and the corresponding NPR report) neglect one major factor. In general, the criticisms leveled by Vul et al are valid. However, they are problems of poor experimentation. There is nothing about fMRI that inherently causes these problems, generally they arise from scientists either not being careful or being overzealous in their claims. The same criticisms can be leveled against any technique in any field. The only difference is that fMRI results in pretty pictures.
I have my own problems with the social neurosciences (mostly variations on "Who cares what discrimination looks like in the brain anyways? What does that tell us about disrimination or the brain?) But the problems of fMRI are like the problems in ERP, TMS, PET, Western Blots, Southern Blots, Electrophloretic Gels, and any other technique you can think of, they are due to incorrect interpretations of the data rather than a problem with the data. Many of Vul's examples of bad fMRI studies are examples of bad studies, studies making claims well beyond the extant data. The fact that fMRI was is really a non-factor.