To get something similar to CONN's first-level previsualization in a nifti mask: get the BETA*.nii file as pointed by Alfonso, then use SPM ImageCalc (or something similar) to filter the voxels. Use two filters:

1. generate positive mask: use equation (i1.*(i1>0.25)) to get the positive thresholded mask (with similar threshold as CONN, can Alfonso confirm if this is how CONN calculates the thresholded map?)

2. generate the negative mask: use equation (i1.*(i1<0.25))

Bonus: if you have a viewer that supports a bidirectional color scheme (with positive and negative colors), then you can generate one combined mask with one equation: (i1.*(abs(i1)>0.25))

Then you can use these two masks as overlays on your subject's normalized structural image in your favorite nifti viewer.

But it would be great to have this implemented in CONN directly! We also often want to explore (and sometimes even publish) 1st-level results. Isn't it possible to port the 2nd-level visualizations on 1st-level? At least the brain surface viewer? That would be awesome!