Hello Andrew, It's well worth trying to rerun with downsampled images, please let me know if the analysis still fails for the smaller data set. Many Regards Matthew > We have just re-run this on another 64-bit machine with 16Gb of RAM, > 32Gb of swap memory and 8 cores. > It seems to terminate at an even earlier stage of the process and > does not completely load into memory (although it is remotely > submitted job and the logfile does not give many details about the > nature of the error). > > The images are 1x1x1- should we perhaps down-sample the images to > 2x2x2? > The original smoothed gm images were obtained in SPM5- and converted > to nifti using fslchfiletype. They display normally in fslview > however and in fslinfo they are the same dimensions and type as the > mask (data_type:float32 datatype:16). > > Is there anything else we could be doing wrong? We have used > randomise and tbss several times for dti data without running into a > problem like this. > > Thanks > Andrew >