It looks like your definition of a word boundary differ from Perl's. For Perl a word is [a-zA-Z0-9_], so the . is not a word character, so the regexp engine finds a word boundary between error and .

If you want to match "This is an error." but not "error.xls.txt" you're going to have to code it yourself.

I don't see any sure way to do this but /\berror\b(?!\S\w)/ (error, then a word boundary then a non space character then a word character) should work _most_ of the time.

I am not sure I would trust a program here, if you really want 100% accuracy you're probably better off finding a way to log and check the matches, if only so that you can fix your regexp when you find out it failed.

You need to read perldoc perlre more carefully. `error.xls.txt' definately does match /\berror\b/i because a word boundary (as matched by \b) is a transition between a character that matches \w and something that matches \W. You've got an impedence mismatch between what you're considering a word and word boundary and what the regexp engine thinks they are. In this specific instance you probably could catch it with /\berror\b(?!\.\w+)/i, which would rule out error followed by `.extension' or `.txt' (as examples).