I never liked the idea of the Bayesian spam filtering that is most popular with big e-mail providers. I figure, as a software engineer, I have e-mail conversations about spam that might be caught by the filters. And in the meantime, the spammers are including paragraphs of text out of The Hobbit in their messages. Yeah, I know the spam filtering tech has gotten a lot more accurate over the years. But...
Years ago, I decided to do it myself. I set up my own e-mail server and built custom filters. I look for connections that violate SMTP protocols. I look for connections that don't have host names. I look for messages with forged headers or that violate SPF or that come from domains known to use Domain Keys, but don't have one.
I use white lists and black lists and thresholds for things that aren't automatically spam, but if there are enough of them, then it gets flagged.
Sure, I'm doing a lot more work than you are, but I successfully block thousands of spam messages each day, and only very rarely have one sneak through (and then I figure out how it got through, and update my system), and also fairly rarely I'll block a legitimate message (and most of the time that's because of a mail server configuration issue on their end).
Sometimes, it's good to be a geek.