...

But, as I mentioned above, the whole matached string, namely #P{x}, will be replaced once the match is found. So, the character [^#] will also be replaced, that is not what I want.

Try a negative lookbehind: "(?<!#)#" will match any single "#" that is *not* preceded by another #. The previous character (whatever it is) >does not form part of the result.

John.

Thank you so much. It's amazing! I see that regex is really a powerful tool! BTW, in order to replace the pattern #P and avoid the string start with double #, I find it is all right to specify the exact times in matching when using replace_all_regex, for example replace_all_regex( source, boost::regex("([#]\{1\}P)|([#]\{1\}G)"), string("(?1 FIRST)(?2 SECOND)"), format_all ); of course, using the "negative lookbehind" approach also works in this problem replace_all_regex( source, boost::regex("((?<!#)#P)|((?<!#)#G)"), string("(?1 FIRST)(?2 SECOND)"), format_all ); For my case, the program might be used to handle a LARGE text (perhaps several megabytes). So, I am wondering which approach is faster! Thanks again!