The odd thing is that it crashes 85% of the way through. Are these trials randomized? If so, that would eliminate a particular stimulus as being problemmatic and points more towards a resource issue. If you look at the data files on the crash machines, are you getting data or are they blank files? Also, check the exact version of the Windows Media PLayer that is running on these machines--I'd just be curious to know if the working machine has an older version maybe. Finally, can you describe the typical crash in any detail?