When I first looked at the TIOBE methodology some years ago, all they did was searching google, yahoo and the microsoft search engine (precursor to bing) for "$language programming" and counted the matches, and did some kind of averaging.

Needless to say that this correlates only very, very weakly with the actual popularity of the language.

I believe that TIOBE now uses more sources, but IMHO the stigma of unscientific approach and FUD still sticks to their "data".