Just posted on slashdot: a Bayesian approach to the problem that reports to have rates of 0.5% on false positives and 0% false negative: http://www.paulgraham.com/spam.html -- Frank