Very good points smile Character classes are indeed faster than alternating subpatterns, as long as the two patterns are comparable in length (remember that mirc has to parse the string before passing it to PCRE). I guess the fastest alternative would be ([^a-z\d\s]*)