Occasionally I'll run into the situation where fixing a bug requires that I delete a section of code. The TDD purist would (I assume) advocate writing a failing test, deleting the code, then watching the test pass.
Now, it seems really strange to have a test asserting that some code was removed. Sure, I suppose it would ensure no one dug into source control and put that code back in, but is it worth it? If it is worth it, it certainly seems less valuable than writing a test for code that has been added, right?