I was intrigued by the use of searchlight analysis by Dirk Walther et al. in "Natural Scene Categories Revealed in Distributed Patterns of Activity in the Human Brain." The searchlight analysis was not the primary interest but rather a control.
The main analysis was ROI-based, with both functional and anatomical ROIs. Much of the paper describes and interprets the accuracy of these ROIs, but they also wanted to "explore brain regions outside of our predefined ROIs," which they did with the searchlight analysis.
The authors ran a searchlight analysis in each subject, with the searchlight sized to approximate the number of voxels in the ROIs. Matching the searchlight and ROI size improves interpretability, given how much spatial structure there is in fMRI data. It's obviously an imprecise matching, since the ROIs followed anatomical and/or functional boundaries and the searchlights did not, but it is a start.
Reading the paper, I got the impression that it could have stood on the ROI-based analysis alone, but the authors (or reviewers) asked if information (as they measured it) was present in other areas. The searchlight analysis was useful for answering that question; it would certainly have been worrying if voxels throughout the brain showed the same classification characteristics as the ROIs. The searchlight analysis also served as a sort of sensitivity test: accurate searchlights were found near the ROIs, despite none of the searchlights exactly matching the shape of the ROIs. This indicates that the classifications found in those areas were not some fluke of the exact voxels picked but rather present in multiple subsets of voxels in those areas.
Walther, D., Caddigan, E., Fei-Fei, L., & Beck, D. (2009). Natural Scene Categories Revealed in Distributed Patterns of Activity in the Human Brain Journal of Neuroscience, 29 (34), 10573-10581 DOI: 10.1523/JNEUROSCI.0559-09.2009