From FSL FEAT user guide:
Thresholding: After carrying out the initial statistical test, the resulting Z statistic image is then normally thresholded to show which voxels or clusters of voxels are activated at a particular significance level.
If Cluster thresholding is selected, a Z statistic threshold is used to define contiguous clusters. Then each cluster's estimated significance level (from GRF-theory) is compared with the cluster probability threshold. Significant clusters are then used to mask the original Z statistic image for later production of colour blobs. This method of thresholding is an alternative to Voxel-based correction, and is normally more sensitive to activation. You may well want to increase the cluster creation Z threshold if you have high levels of activation.
The FEAT web page report includes a table of cluster details, viewed by clicking on the relevant colour-overlay image. Note that cluster p-values are not given for contrasts where post-threshold contrast masking (see below) is applied, as there is not a sensible p-value associated with the new clusters formed after masking.
If Voxel thresholding is selected, GRF-theory-based maximum height thresholding is carried out, with thresholding at the level set, using one-tailed testing. This test is less overly-conservative than Bonferroni correction.
You can also choose to simply threshold the uncorrected Z statistic values, or apply no thresholding at all.
http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FEAT/UserGuide
From FIRST user guide:
randomise has the following thresholding/output options:
Voxel-based thresholding, both uncorrected and corrected for multiple comparisons by using the null distribution of the max (across the image) voxelwise test statistic. Uncorrected outputs are: <output>_vox_p_tstat / <output>_vox_p_fstat. Corrected outputs are: <output>_vox_corrp_tstat / <output>_vox_corrp_fstat. To use this option, use -x.
TFCE (Threshold-Free Cluster Enhancement) is a new method for finding "clusters" in your data without having to define clusters in a binary way. Cluster-like structures are enhanced but the image remains fundamentally voxelwise; you can use the -tfceoption in fslmaths to test this on an existing stats image. See the TFCE research page for more information. The "E", "H" and neighbourhood-connectivity parameters have been optimised and should be left unchanged. These optimisations are different for different "dimensionality" of your data; for normal, 3D data (such as in an FSL-VBM analysis), you should just just the -T option, while for TBSS analyses (that is in effect on the mostly "2D" white matter skeleton), you should use the --T2 option.
Cluster-based thresholding corrected for multiple comparisons by using the null distribution of the max (across the image) cluster size (so passé!): <output>_clustere_corrp_tstat / <output>_clustere_corrp_fstat.
To use this option, use -c <thresh> for t contrasts and -F <thresh> for F contrasts, where the threshold is used to form supra-threshold clusters of voxels.Cluster-based thresholding corrected for multiple comparisons by using the null distribution of the max (across the image) cluster mass: <output>_clusterm_corrp_tstat / <output>_clusterm_corrp_fstat.
To use this option, use -C <thresh> for t contrasts and -S <thresh> for F contrasts.