Yeah I've used FizzBuzz myself, and under interview conditions it's actually much harder to implement a correct (for edge cases, etc.) solution in a short time.

Yeah I've used FizzBuzz myself, and under interview conditions it's actually much harder to implement a correct (for edge cases, etc.) solution in a short time. I always took that into account. However I also added the twist that I wanted the candidates to also write the tests for the program in TDD style. I hoped that this would actually help, by asking the interviewee to follow a repeatable pattern of first expressing what they wanted to code to do and then implementing. We also did ping-pong pairing so I could help write a test, or do an implementation if things were going poorly, to see if that would jump start the candidates brain into the problem. The results were quite frankly a disaster, no better than what Joey found (though I didn't do enough interviews for a statistical sample). Unfortunately my limited observations have lead me to believe that too many developers (particularly in the enterprise environments) are used to working on legacy codebases that they hack and change until the bug they are fixing goes away. Many rarely implement features much, if at all, and the ones they do are heavily broken.