On Jun 15, 2007, at 9:06 AM, hamann.w@t-online.de wrote:
> A simpler approach might be to blacklist senders that try multiple
> non-existent recipients,
> regardless of mx priority

In Postfix I tarpit after the first bad recipient and eventually
disconnect. That's cut things down quite a bit.

> BTW: at one time I was quite happy with some pre-filtering on my
> private mail (which is
> fetchmail ultimately feeding to SA) until I found that SA would no
> longer recognize some
> spam in the bayes section. So, if capacity permits, it might be a
> good idea to feed (a random
> sampling of) pre-filtered spam to sa-learn

I have a few spamtrap addresses that feed directly to sa-learn.
Seems to work pretty well.

Now to deal with the companies that send out billing, etc. through a
third party that uses the original company's return address but third-
party servers. I even had to explain SPF to an anti-virus company,
not sure it they got it.