There are two basic causes of poor accuracy, setup problems and improper training:
SpamSieve itself may be working properly, but if it doesn’t have access to your incoming messages, it won’t be able to tell which ones are spam. First, check SpamSieve’s log to see if SpamSieve is analyzing your messages or if you need to adjust its preferences. If SpamSieve is not analyzing your messages, you can check its setup by following these steps:
The setup processes for Apple Mail, Entourage, and PowerMail involves creating a rule in the mail program. When this rule is applied to a message, SpamSieve can examine its contents and send it to the spam folder (if it’s spam). Make sure that you have a rule that looks exactly like the one in the SpamSieve manual.
To test that the rule works, select a spam message in your mail program. If you are using PowerMail, go to the Mark as Spam pane of the preferences and temporarily uncheck mark it as good at the bottom of the window. Use the Train Spam (Apple Mail or Entourage) or Mark as Spam (PowerMail) command to tell SpamSieve that it is spam. Drag this message to your inbox and select it again. Then manually apply the rule:
If the rule worked, then the problem is that another rule (above it in the list) is preventing the SpamSieve rule from being applied. Dragging the SpamSieve rule higher in the list should help, as this gives the SpamSieve rule priority. In Entourage, Mailing List Manager rules are implicitly higher than the regular rules, and they can often interfere with the regular rules, even if it looks like they wouldn’t. Try unchecking Do not apply rules to list messages in the Advanced tab of each Mailing List Manager rule, or delete the Mailing List Manager rules and re-create them as regular rules.
If the rule did not work, see below for how to fix it:
The filter might be configured such that it excludes most messages from spam filtering. Try editing the Spam: evaluate filter so that it has just a single condition that says Always.
Quit PowerMail, then delete the folder:
/Users/<username>/Library/Application Support/PowerMail
and then re-launch it. (If the script files that PowerMail uses to communicate with SpamSieve may be damaged, this will fix them.)
SpamSieve is nearly 100% accurate, but only when properly trained. For best results, the corpus should have about 65% spam messages, as shown at the bottom of the Statistics window. You can manually train SpamSieve with more messages to improve this ratio, or let it auto-train itself with incoming messages.
The messages in the corpus should be representative of the messages that you receive. Adding more messages to the corpus generally improves accuracy, but it is not necessary to have more than a few thousand messages in the corpus (and having more than that will probably reduce accuracy). If the corpus is very large or very unbalanced, and there are so many messages in it that you cannot get it close to the 65% ratio, then you should reset the corpus and re-train SpamSieve. For more information, see this section.