Real hit count with proxy

[Ahmet]

Hi,
I am trying to count web hits of users. With using proxy it is seems to
be easy (I tried squid, tiny, privoxy).
But it is obvious that the hits in the logs not purely the hits that
users wanted to do.
For example when a user goes to cnn.com, cnn.com calls other ad pages or
non-ad pages. But because that think happens at users browser it is seen
as an user hit in logs. So for real hit count an analysis must be made
on logs.
Do you know any tool, proxy that can help in such analysis?

Second choice is writing own tool that can be parsing the logs and doing
an analysis on referer field. But in that case (assuming cnn.com
referred pages as non hit) users clicking on a link on the page
(cnn.com) can not be distinguished.
To further investigate the issue I listened (by ethereal) outgoing
packets for a usual user behavior (clicking on a link) and  web site
called web pages. They all seem to have same headers and similar header
values in request packets. So I stucked and could not found any possible
piece of evidence to track and distinguish the behaviours.
Is there a known theoretical or practical way for distinguishing this
behaviours?

Not: I know this mailing list is not the most appropriate list to ask
this question. But any help will be gratefull.

AB