Junk scores

Mail's junk filer is rule based, not score based. Basically, any message that meets all the rule criteria is considered junk -- IOW, it is an all or nothing system.

You can modify the junk rule just like any other: go to Mail Prefs -> Junk -> Advanced button & add filter criteria or change its action behavior.

More info on this can be found in Mail's Help topics under the name "Changing the junk mail filter."

That Mail's junk filter is rule-based does not mean there is not a threshold that causes it to mark messages as junk or not.

If you go to Mail > Preferences > Junk Mail and click the Advanced button, you'll see that one of the conditions in the rule that determines the behavior of the junk filter is Message is Junk Mail. The junk filter evaluates that condition by statistical means using the data stored in the LSMMap2 file. Actually, that's precisely the reason you can improve the filter accuracy by training it. Training would be impossible otherwise.

To answer the OP's question, this feature is missing from Mail, indeed, although I believe this is the first time I've seen someone complaining about that...

That Mail's junk filter is rule-based does not mean
there is not a threshold that causes it to mark
messages as junk or not.

Well, yes & no. What I guess I should have said is that while one of the rules ("Message is Junk Mail") has a threshold, the filter, at least in the default state, requires that all of the rules evaluate to true or the Junk Filter rule does not take whatever action it is set to take.

IOW, for the OP's problem, he can still add a decision to the rule (typically of the "not" form) that would exclude particular emails from the filter's action, no matter how the "Message is Junk Mail" decision rated it.

One of the criteria in preferences:advanced is
"Message is junk mail." I imagine this represents
Mail's trainable junk filter.

Yes, this is what the help topic means when it states, "To use the junk mail database when identifying junk mail, make sure "Message is Junk Mail" is one of the conditions (it's specified by default)."

Also note that you can continue to "train" Mail's junk mail database even if you don't use it as a decision criteria for whatever action you set the filter to take.

Overall, I think Eudora's junk score scheme is more
satisfactory, although more and more junk is getting
through, due to the spammers' new tricks. Maybe it's
time to try Thunderbird.

A trainable filter is more difficult for spammers to fool, since it relies in part on your personal decisions for its analysis. When incorporated into a rules-based system, it becomes a pretty powerful way to detect spam without mistaking non-spam as something to act on.

For instance, I've added a decision criterion to exclude any message that contains the name of the company I work for, so even if a message is from an unknown sender I've never replied to, it won't be acted on by the filter if it contains that text.

Give training the junk filter a try before making any conclusions -- you can do this while in Automatic mode, no need to switch to Training mode if you don't want to.

As I suggested earlier, if you can determine something distinctive about those incorrectly identified important messages, you can also add custom decision criteria to prevent the filter from applying its action(s) to them. It is hard to say what would work for you, but here are some possibilities:

• Message content does not contain some distinctive phrase such as company or project name; or some phrase you mention on a web site or in correspondence, church bulletins, newsletters, or similar things where people would find your email address. You may wish to render the phrase as an image on web sites for added protection against spammers.

• Any recipient does not contain the name of some member of a group who would receive the same mass mailings as you do.

• From does not contain some domain name or part thereof that is particular to a group or organization you belong to or you expect to get email from.

• Any other header info does not contain some distinctive text. Note that you are not limited to the four default choices (From, To, CC, Subject). You can add any header via the "Edit header list..." item at the bottom of the popup, for example "Content-Type" or "X-Mailer." (To see the possibilities, open one of the mis-identfied messages & choose View -> Message -> Long Headers.)

•• Don't overlook the possibility that it is your ISP's junk filter that is mis-identifying messages, which will 'train' Mail's filter to do the same (I think). If this is the case, you can see if your ISP offers any way to 'tune' its filter to better suit your needs, or even tell Mail not to use that as a criterion in its own filter.

What's particularly nifty about this approach is you can use it in conjunction with other rules to further weed out potential spam, allowing you to use more liberal criteria in the Junk Mail filter itself & still mark or move the suspect messages as you like.

In this case, you could create multiple rules to apply the reverse of the above, changing the "does not" form to the 'does' form, allowing you to move suspect items to one or several folders for further examination.

Of course, this takes some work on your part, but it gives you extensive control of spam filtering that is hard to beat.

Thank you! I started off thinking I would just dash off an example or two but as I did so, I realized there are several possibilities I hadn't considered before that should be useful to me as well.