I had a problem with  a higher level analysis:

    An exception has been thrown
    Runtime error:- detected by Newmat: matrix is singular

Searching the mailing list gave me a possible solution, that this could 
happen, if the mask includes voxels in the filtered_func_data 4D file 
with all zeros. In a post 
about a year ago it was stated that this problem is solved with a new 
FSL version. However, recently 
there was a comparable discussion with the result that the mask should 
not include voxels with all zero. So, is it still a problem when the 
mask includes zero-voxels? If so, what is the best trick to avoid the 
creation of a mask as preparatory step in the higher level analysis and 
instead use an alternative a mask similar as one can do in the first 
level analysis at the end of the .fsf-file? So far I copied a custom 
made mask into the gfeat-directory at the beginning of the analysis and 
made sure that it is write protected. Did work fine until today...

Thanks for the help,