K9 by Robin Keir is a Bayesian filter & learns what you consider valid & spam. And when spammers changed their techniques as they have done on the 45 day sample it adapts. This sample is results from my use of K9 over a 215 day period. The most important statistic to me is "Percentage emails misidentified as Spam (false positives)." This is the number of good emails that K9 has marked as spam. For me anything above 0.0% is unacceptable. I don't want to lose one email. K9 seems to have learned what I consider good email because it has misidentified 21 emails in the past 165 days none of those have been in the past 48 days.

K9 Spam Filter Statistics Column 2: Since Sunday, July 13, 2003 3:01:19 PM (215 days) Column 1: Since Wednesday, December 10, 2003 6:07:55 PM (65 days)   Total number of emails processed 4,077 14,839 Number of emails processed per day 63/day 69/day Number of Good emails processed 2,214 6,272 Number of Good emails processed 54.3% 42.3% Number of Spam emails processed 1,863 8,567 Number of Spam emails processed 45.7% 57.7% Percentage of emails that matched whitelist rules 9.0% 8.8% Percentage of emails that matched blacklist rules 0.0% 0.0% Number of emails re-classified to Good 0 21 Number of emails re-classified to Spam 95 285 Percentage emails misidentified as Spam (false positives) 0.0% 0.1% Percentage emails misidentified as Good (false negatives) 2.3% 1.9% Overall accuracy 97.7% 97.9% Keir.net K9 Page: http://keir.net/k9.html