The first question that data miners must ask about counterterrorism is: Will terrorist groups look different from other groups in the sort of datasets that could plausibly be collected. I will suggest several reasons to suppose that the answer is yes. The second question, therefore, is: Can data-mining techniques distinguish terrorist groups, even though they are trying hard to be indistinguishable. Again, the answer seems to be yes. I will outline the application of matrix decomposition techniques to datasets representative of travel data, communication data, and internal structure of terorrist organisations.