This perhaps of interest to those of us who worry about such things:

Henson, R. (2005). What can functional neuroimaging tell the experimental psychologist? The Quarterly Journal of Experimental Psychology, 58A(2), 193?233.

abstract
I argue here that functional neuroimaging data?which I restrict to the haemodynamic techniques of fMRI and PET?can inform psychological theorizing, provided one assumes a ?systematic? function?structure mapping in the brain. In this case, imaging data simply comprise another dependent variable, along with behavioural data, that can be used to test competing theories. In particular, I distinguish two types of inference: function-to-structure deduction and structure-to-function induction. With the former inference, a qualitatively different pattern of activity over the brain under two experimental conditions implies at least one different function associated with changes in the independent variable. With the second type of inference, activity of the same brain region(s) under two conditions implies a common function, possibly not predicted a priori. I illustrate these inferences with imaging studies of recognition memory, short-term memory, and repetition priming. I then consider in greater detail what is meant by a ?systematic? function?structure mapping and argue that, particularly for structure-tofunction induction, this entails a one-to-one mapping between functional and structural units, although the structural unit may be a network of interacting regions and care must be taken over the appropriate level of functional/structural abstraction. Nonetheless, the assumption of a systematic function?structure mapping is a ?working hypothesis? that, in common with other scientific fields, cannot be proved on independent grounds and is probably best evaluated by the success of the enterprise as a whole. I also consider statistical issues such as the definition of a qualitative difference and methodological issues such as the relationship between imaging and behavioural data. I finish by reviewing various objections to neuroimaging, including neophrenology, functionalism, and equipotentiality, and by observing some criticisms of current practice in the imaging literature.

In which this pleasing analogy is noted:

?the use of functional imaging to understand the brain? [is like] ?trying to understand how a car engine works, using only a thermal sensor on a geostationary satellite? (original source unknown; apologies for plagiarism)

Henson is not convinced. Or to put it another way, he is convinced of the utility of neuroimaging for psychologists. It’s an interesting, and almost conversational, read. I suspect that the ‘systemmatic function?structure mapping’ assumption is probably like the adaptationist position in evolutionary biology. You can’t prove it, you’re certain it must sometimes be wrong and misleading, but it does useful work for you so you might as well use it.

One ‘best-practice’ caveat the paper mentions about imaging is

…a minimal requirement for deducing the presence of a different function (F2) is an interaction in which one region shows a reliably greater change in activity across conditions than at least one other region.

Which, I think, is saying that if you have a notional function (which you hope is involved in your challenge task but not in your baseline task) then you do not demonstrate it (or localise it) by selecting a region which survived your SPM statistical tests of difference. You’ve just found a region which responds more in at least this one task. Henson (i think) is saying that you need to include region as an (independent) variable of analysis and show that there are tasks which increase activation in region A more than in region B, and vice versa.