SpamAssassin Score Explained: What It Means and How to Lower It

deliverabilitySpamAssassin Score Explained: What It Means and How to Lower It

A SpamAssassin score is the total of the points from every test a message triggers. By default, a message that scores 5.0 or more is tagged as spam. Lower is better, and the score can even be negative.

This guide is for senders who got a SpamAssassin score from a test tool or found one in a recipient's headers, and want to know what it means and what to fix.

Short answer

  1. Find X-Spam-Status in the raw headers of a message that went through SpamAssassin.
  2. Compare score= with required=. At or above the required score, the message is tagged as spam.
  3. Read tests= to see which rules added (or took away) points.
  4. Fix what you control: authentication, message headers and content, and the quality of your list.
  5. Send again and retest, one change at a time.

What SpamAssassin is (and who uses it)

Apache SpamAssassin is an open-source project of the Apache Software Foundation. Its site calls it "the #1 Open Source anti-spam platform giving system administrators a filter to classify email and block spam". It runs many heuristic and statistical tests on each message: text analysis, Bayesian filtering, DNS blocklists and collaborative filtering. As of November 2026, the current release is 4.0.2.

The key point for senders: SpamAssassin is a filter that system administrators install and configure on their own mail servers. The spam threshold, the rules and the point values are their choice, so the same message can score differently from one server to the next.

How a SpamAssassin score adds up

Every SpamAssassin rule has a point value. When a message matches a rule, its points are added to the total. Points can be positive (looks like spam) or negative (looks legitimate). At the end, SpamAssassin compares the total with the server's threshold, required_score.

Illustration: blocks of different sizes stack up toward a threshold ribbon, while two weights under the shelf pull the stack down

A few details explain why scores vary:

  • Four score sets. Many rules have four point values, one for each combination of two features: Bayesian filtering on or off, and network tests (such as DNS blocklist lookups) on or off. Set 0 has both off, set 1 has network tests on, set 2 has Bayes on, and set 3 has both on. The server uses the set that matches its setup.
  • A score of 0 turns a rule off. Administrators can disable a rule this way, or raise and lower any rule's points in their local configuration (local.cf).
  • The default values are generated, not hand-picked. The project produces them with mass-check scripts and a perceptron, a training process run over large collections of spam and non-spam mail. According to the default scores file, they are weighted to produce "roughly 1 false positive in 2500 non-spam messages using the default threshold of 5.0".
  • Values change with updates. The scores ship in a rules file that sa-update replaces, so any table of rule points, including the one below, is a snapshot.

Why scores can be negative

Rules that point to legitimate mail subtract points. A valid DKIM signature (DKIM_VALID) is worth -0.1 by default, and so is a valid signature from the author's domain (DKIM_VALID_AU). When Bayesian filtering is on (sets 2 and 3), BAYES_00, which fires when the Bayes filter is almost sure the message is not spam, subtracts 1.5 or 1.9 points depending on the score set.

If the negative rules outweigh the positive ones, the total drops below zero. A negative score just means the message looks very unlike spam to that server.

The threshold: required_score

The threshold is set with required_score, and the default is 5.0. Apache's own documentation calls 5.0 "quite aggressive" and says it is suitable for a single-user setup. It suggests that ISPs "should probably set the default to be more conservative, like 8.0 or 10.0".

So whether a message counts as spam depends on the server's threshold, not on a universal line. A score of 6.2 means the message is 1.2 points over the default threshold. It does not mean "62% spam", and on a server set to 8.0 the same message would pass.

How to see your score

SpamAssassin adds its result to the message headers. In most email clients, look for an option like "Show original", "View source" or "View message headers" to see them. The default headers are:

  • X-Spam-Flag: YES when the message is over the threshold.
  • X-Spam-Status: Yes or No, then score= (the total), required= (the threshold), tests= (the rules that matched), autolearn= and version=.
  • X-Spam-Level: one * per point.
  • X-Spam-Checker-Version: which SpamAssassin version checked the message.

Here is an illustrative example. The domain and values are made up, not real output from any server:

X-Spam-Flag: YES
X-Spam-Level: *******
X-Spam-Status: Yes, score=7.3 required=5.0 tests=DKIM_INVALID,DKIM_SIGNED,
    DMARC_NONE,RCVD_IN_PBL,SPF_SOFTFAIL,URIBL_BLACK autolearn=no
    version=4.0.2

How to read it: the message scored 7.3 against a threshold of 5.0, so this server tagged it as spam. Assume the server uses score set 1 (network tests on, Bayes off), which is also why no BAYES_ rule appears in tests=. With the default points for that set, the DKIM signature didn't verify (DKIM_SIGNED and DKIM_INVALID, 0.1 each) and SPF soft-failed (SPF_SOFTFAIL, 0.972). Because neither SPF nor DKIM passed for the From domain, DMARC failed, and the domain's policy is p=none (DMARC_NONE, 0.898). The authentication failures (DKIM_INVALID, SPF_SOFTFAIL, DMARC_NONE) add up to about 2 points. The rest came from the sending IP being on a blocklist (RCVD_IN_PBL, 3.558) and a link to a listed domain (URIBL_BLACK, 1.7), for a total of 7.328.

Fixing authentication alone removes those failures and adds the small negative rules (DKIM_VALID and DKIM_VALID_AU at -0.1 each, SPF_PASS and DMARC_PASS at -0.001 each), while DKIM_SIGNED stays. That brings the total to about 5.2, still just over the line, because the blocklist and link rules carry most of the weight.

The header names above are the defaults, and administrators can change them. With the default report_safe setting, a message over the threshold is also wrapped in a new report message; with report_safe 0, SpamAssassin only adds the headers.

Many test tools show their own scale, such as 0-10 or 0-100. Those are the tool's scores, not SpamAssassin scores, even if SpamAssassin is one of their inputs. Look for the X-Spam-Status line or the tool's SpamAssassin section to compare like with like.

What is a good SpamAssassin score?

A good SpamAssassin score is one below the receiving server's required score, with room to spare. Since you rarely know that server's threshold, a buffer below the default 5.0 is sensible: some servers set a lower threshold than the default.

Vendors suggest different buffers. MailerCheck recommends aiming for 0 to 2 and staying under 4, and SyncGTM targets 2.0 or lower for cold email. These are their guidelines, not SpamAssassin rules.

A bad SpamAssassin score is one at or above the threshold, so 5.0 or more on a default setup. Anything close to it is a warning: one more rule and the message tips over.

Even a very low score does not guarantee the inbox, as the section on inbox placement below explains.

Common rules and what they add

These are default points from Apache's 50_scores.cf in the development trunk, checked on November 16, 2026. Your installation may use different values after sa-update, and administrators can override any of them. The two point columns split the four score sets by network tests: on (sets 1 and 3) or off (sets 0 and 2). Where a cell shows a range, the value also depends on whether Bayes is on in that set.

Rule What it flags Network tests on (sets 1, 3) Network tests off (sets 0, 2)
SPF_PASS SPF passed -0.001 -0.001
SPF_FAIL SPF failed (-all) 0.919 (set 1), 0.001 (set 3) 0
SPF_SOFTFAIL SPF soft-failed (~all) 0.665 to 0.972 0
DKIM_SIGNED Message has a DKIM signature 0.1 0.1
DKIM_VALID DKIM signature verifies -0.1 -0.1
DKIM_VALID_AU Valid signature from the author's (From) domain -0.1 -0.1
DKIM_INVALID DKIM signature present but doesn't verify 0.1 0.1
DMARC_PASS DMARC passed -0.001 -0.001
DMARC_NONE DMARC failed, domain policy is p=none 0.898 0.001
DMARC_QUAR DMARC failed, domain policy is p=quarantine 1.198 0.001
DMARC_REJECT DMARC failed, domain policy is p=reject 1.797 0.001
RCVD_IN_PBL Sent from an IP address on the Spamhaus PBL 3.335 to 3.558 0
URIBL_BLACK Contains a link to a domain on the URIBL blocklist 1.7 0
BAYES_99 Bayes filter is almost certain it's spam 3.5 with Bayes on (set 3), 0 without 3.8 with Bayes on (set 2), 0 without
MISSING_DATE No Date header 1.36 to 1.396 1.8 to 2.739
HTML_IMAGE_ONLY_24 HTML part is mostly images with very little text 1.282 to 1.618 1.328 to 2.799

Two things stand out. Authentication rules carry small points on their own, but failures add up, and a signed message with a valid signature nets out at zero or below (DKIM_SIGNED +0.1, DKIM_VALID -0.1, DKIM_VALID_AU -0.1). The heaviest rules are about reputation and content: blocklisted IPs and links, and the Bayes filter.

How to lower a high score

Start with the parts you control, roughly in this order:

  1. Authenticate your domain. Publish SPF, sign with DKIM using your own domain, and add a DMARC record. That removes the SPF, DKIM and DMARC failure rules and adds the small negative ones. Our guide to how SPF, DKIM and DMARC work together covers the setup, and the free SPF, DKIM and DMARC checker shows whether the records are in place.
  2. Fix headers and message structure. Make sure every message has a Date and Message-ID header, and send HTML email with a plain-text part. Check this especially if you send from your own code rather than an email service provider.
  3. Clean up the content. Avoid image-only HTML, add real text next to images, skip all-caps "shouting" subjects, and don't use link shorteners or link to domains that are on blocklists.
  4. Check blocklists and sending reputation. If tests= shows rules like RCVD_IN_PBL or URIBL_BLACK, look up the listed IP or domain with the blocklist operator. New domains and IPs need a gradual start, which our email warmup guide explains. If you're already in trouble, follow the steps to recover your sender reputation.
  5. Keep your list clean. SpamAssassin doesn't look at your list, but bounces and spam traps damage the reputation that blocklists and mailbox providers judge you by. Clean your email list before big sends, and reduce bounce rates over time.

Change one thing at a time and retest, so you know which fix moved the score.

A low score is not inbox placement

A SpamAssassin score tells you how one filter, with one configuration, judged one message. Large mailbox providers such as Gmail, Outlook and Yahoo run their own filtering and weigh signals a test can't see: your sending reputation, spam complaints, bounces and how readers engage with your mail. A message can score well in a test and still land in spam.

Gmail and Yahoo also have their own sender requirements. Bulk senders to Gmail (5,000 or more messages a day) need SPF, DKIM and DMARC, a spam rate under 0.3% and one-click unsubscribe. Yahoo asks bulk senders for SPF and DKIM, DMARC at least at p=none, and a spam rate under 0.3%. For the background, read Gmail's new email delivery standards.

Treat the SpamAssassin score as a useful check for fixable problems, not as a prediction of where your mail lands.

Troubleshooting

Two tools show different scores for the same message. That's normal. Each tool may run a different SpamAssassin version, a different score set (with or without Bayes and network tests), its own threshold and its own custom rules. Compare the tests= lists rather than the totals.

Authentication passes, but the score is still high. Look at the other rules in tests=. Content rules (image-only HTML, missing headers), link blocklists (URIBL_*) and the Bayes rules (BAYES_*) can add more points than authentication takes away.

tests= lists a rule you don't recognize. Search for the rule name in Apache's rules and scores file. The name usually hints at what it checks, and the file shows its default points.

FAQ

What is a good SpamAssassin score?

Anything below the receiving server's required score, ideally with a buffer. The default threshold is 5.0; vendors such as MailerCheck suggest aiming for 0 to 2.

Is a score of 5 spam?

On a server with the default settings, yes: a message is tagged as spam at 5.0 or more. Servers with a higher threshold, such as 8.0 or 10.0, would let it through.

Can a SpamAssassin score be negative?

Yes. Rules that indicate legitimate mail, such as a valid DKIM signature or a Bayes result of "almost certainly not spam", subtract points. If they outweigh the rest, the total is negative.

Does a low SpamAssassin score guarantee the inbox?

No. It shows that one filter didn't flag the message. Mailbox providers use their own filters and also weigh your reputation, complaints and bounces.

Who sets the SpamAssassin threshold?

The administrator of the mail server that runs SpamAssassin, with required_score. The default is 5.0, and Apache suggests 8.0 or 10.0 for ISPs.

Next step: fix the part you can fix in DNS

SpamAssassin's authentication rules are the part of the score you can fix in DNS. Check your domain with the free SPF, DKIM and DMARC checker: enter your domain and click Check Domain. You get a Pass, Warning or Fail card for each record, an overall grade from A to F with a score out of 100, and a list of recommendations. That grade rates your domain setup; it isn't a SpamAssassin score. The checker doesn't check blacklists or spam reputation.

Once your domain authenticates, verify the list before the next send. Try the free email verifier on a single address, or sign up to clean a whole list with 200 free validations, no credit card required.

Join 1,000+ CompaniesImproving Email Deliverability

Start with 200 free validations. Upgrade only when you're ready.

No credit card required • Cancel anytime