with ssi from an shtml file:

<!--#include virtual="/cgi-bin/sinister_tracking_script.pl"-->
[download]

and in sinister_tracking_script.pl:

my $page = $ENV{HTTP_REFERER}; # iirc
[download]

or more simply (unless you want to take the opportunity to generate the html for some of the page) and without requiring SSI:

<img src="/cgi-bin/creepy_bug.pl">
[download]

and from creepy_bug.pl you can retrieve the page address in the same way, log it and then either return a standard image file, print a counter for old times sake, or just not bother. Note that this method is vulnerable to browser caches and proxies and so on: best to add a unique identifier to the image address each time, so that the browser definitely asks for it. and yes, you can set a cookie while returning an image file, which you can't with ssi-included html.

Don't use this technique on remote pages, btw, especially not with very small images. People will make wild accusations.

ps. google is your friend.