Jonathan Corbet, the Grumpy Editor over at LWN.net, has a reasonably good review of the current offerings in the world of Bayesian spam email filters. His tests hinted that SpamAssassin remains difficult to beat in terms of accuracy, but that it’s still the slowest and most computing-intensive of all the solutions out there (in part because SpamAssassin does a lot more than just act as a Bayesian filter). It’s the system I run all my incoming mail through, but I definitely feel the processor crunch at times — and it’s definitely the kind of service I’d love to offload if there were a reasonable and inexpensive way to do so.

(For those of you who aren’t hip to the lingo of internet system administration, Bayesian spam filters are “trainable” applications that scan incoming email and make predictions about whether any given message is spam, predictions that are based in part on the content of prior legitimate and illegitimate messages to the same users. Back in 2002, Paul Graham wrote an article which posited that applying Bayesian probability theories to email might help alleviate the growing spam problem, and since then the notion has established itself as one of the cornerstones of email administration.)