Mortgage Credit Score: Why Mine Changed 4 Ways

My credit app showed 741 all year. But when I applied for a credit card, financed a car, and took out a mortgage, none of those lenders saw 741.

The card issuer scored me at seven hundred and thirty-six. The auto lender came back with seven hundred and eighty-two, which is not a typo and not a gain. The mortgage lender pulled three scores at once: seven hundred and twenty-eight, seven hundred and thirty-four and seven hundred and thirty-one, and used the middle one.

Three applications produced four numbers from one file in one year, and none was wrong. There is no single “your credit score”: lenders use different models, data, and bureau files for different decisions.

It is the number people anchor on and it is usually the least relevant one.

Free credit apps often show a VantageScore, and some show a FICO score. Both are real scores computed from real data, and a free FICO is not automatically inferior or merely educational. What it may not be is the model, version, bureau or date your lender uses for its decision. Mine showed a VantageScore from a single bureau, refreshed weekly: useful for monitoring, but a poor predictor of what an underwriter will see.

The distinction that matters is between monitoring and underwriting. A monitoring score tells you the direction your file is moving, which is exactly what you want for tracking progress. An underwriting score is what a specific lender purchases for a specific decision, and it may be a different model, a different version of that model, and drawn from a different bureau.

The card issuer pulled a FICO Score 8 from TransUnion and got seven hundred and thirty-six. Five points below my app, which is unremarkable: different model, different bureau, different day.

FICO 8 is widely used across card and general consumer lending, and if there is such a thing as a default score, it is this one. It runs on the familiar three hundred to eight hundred and fifty scale.

This produced the number that confused me most.

Industry-specific FICO scores, including auto and bankcard versions, use a 250–900 range rather than 300–850. A 782 on that scale is not the same as a 782 on the standard scale, so comparing them directly is meaningless.

Within the auto model, a higher score means lower predicted risk, just as it does within FICO 8. But 782 on the auto scale does not necessarily indicate better standing than 741 on another scale: the models predict different outcomes, use different ranges, and may read different bureau files. Once I understood that, the dealer’s quote made sense.

The reason these industry versions exist is that the question being asked is genuinely different. Predicting whether someone will repay a car loan is not the same problem as predicting general credit risk, and a model tuned to the first will weigh an existing auto loan history more heavily than one tuned to the second. That is a legitimate difference rather than marketing.

This is the most structured of the four and works unlike the others.

Mortgage lenders order a tri-merge: a combined report from all three nationwide bureaus, with one score from each. For conventional loans delivered to the GSEs, that has long meant Classic FICO versions. My file was scored that way, although the system is now transitioning. VantageScore 4.0 became eligible for delivery to Fannie Mae in April 2026 through a limited group of approved lenders, and wider access was still being worked out as of this writing. That is a Fannie delivery rule, not a rule for every conventional loan, so ask your lender which model it uses.

One detail that surprised me: the three Classic FICO scores are not one model run three times. Fannie Mae specifies a different bureau-specific version at each: Equifax Beacon 5.0, Experian/Fair Isaac Risk Model V2, and TransUnion FICO Risk Score Classic 04. So a tri-merge gives you three related but distinct models reading three different files.

I got three: 728 from one bureau, 734 from another, and 731 from the third. For a single borrower, the convention is to use the middle score, so my file was underwritten on 731: not the highest or the average, but the median.

• The app number, a VantageScore from one bureau: 741.
• The card pull, FICO 8 from one bureau: 736.
• The auto pull, an industry-specific score on a 250-900 scale: 782.
• The mortgage tri-merge, classic FICO from three bureaus: 728, 734, 731.
• The number actually used for the mortgage decision: 731, the middle of three.

Two things vary at once: each bureau uses its own nonidentical Classic FICO version, and each bureau file can contain different data because furnishers do not all report to all three. The spread reflects both effects, so it is not a pure measure of data differences.

Six points of spread across my three was modest. Larger spreads happen and they usually mean something specific: an account present at one bureau and missing at another, a collection reported to only one, a limit updated at two of the three. That spread is diagnostic. If your three mortgage scores differ by twenty or thirty points, the gap is telling you where to look.

This is the strongest practical argument for checking all three reports rather than one. A single-bureau view is a view of one of three files, and the one you cannot see may be the one your lender ends up using.

The reverse can happen too. If a tri-merge shows one score well below the other two, read that bureau’s file line by line before accepting the middle score as final. The low outlier can reflect an error absent from the other two.

Ask which score the lender pulls and from which bureau. It is a specific question with a specific answer, and in my experience asking it directly gets a direct reply. Knowing the bureau lets you check that report specifically. On freezes, be careful with the obvious shortcut: for a single-bureau product you may only need to thaw one, but a mortgage tri-merge pulls all three, so a partial thaw will stall the application. Ask what the lender needs before deciding what to lift.

Stop comparing numbers across sources. Your app's score this month against your app's score last month is a valid comparison. Your app's score against a dealer's quote is not, particularly if one of them is on the 250-900 scale.

Being near a threshold.

If your app shows seven hundred and forty-one and a lender's cutoff is seven hundred and forty, you are not safely above it. You are near a line, measured by an instrument you cannot see, and the pull that matters could come in five points lower for entirely ordinary reasons.

Leave a margin. If you are close to an important threshold, treat your app score as approximate and build in a buffer rather than applying the week it first crosses the line. That is not a scoring trick; it acknowledges that the score you see is not necessarily the score used.

Next, plan ahead. If a mortgage tri-merge is coming, spend the preceding months lowering balances, avoiding new accounts, and keeping everything current. That matters more than trying to predict which bureau will supply the middle score. You cannot control which score lands in the middle, but you can control what all three files say.

Four scores from one file in one year: 741 in my app, 736 for a card, 782 on an auto-specific scale, and a mortgage decision based on 731, the middle of three scores pulled at once.

None was wrong, and none was the one real score, because no such score exists. Each score is a specific model’s answer to a specific question using the bureau file its buyer selected.

The practical takeaway is simple. Use your app to track a trend, not predict a decision. Ask each lender which score and bureau it uses. Never compare scores from different scales. Near an important threshold, treat the score you can see as having a margin of error.

My four figures are one file over one year, and the mortgage side of this is mid-transition as VantageScore 4.0 becomes available on conventional files. What generalizes is the structure: monitoring scores, general-purpose scores, industry-specific scores on their own scale, and the tri-merge with its middle-score convention.

1. Why is my credit score different from what the lender saw?

Because they are different scores. Free services often show a VantageScore or a FICO score drawn from one or more bureau files, while a lender buys a specific model, version and bureau for its decision, on a specific date. My app showed 741 while a card issuer scored me at 736 and a mortgage lender used 731.

2. What is a FICO Auto Score and why was mine so much higher?

It is an industry-specific version tuned to predict auto loan performance. The industry-specific FICO scores, including auto and bankcard, run on a 250-900 range rather than 300-850, so a higher number there does not translate into a better standing on the standard range. Within the auto model higher still means lower predicted risk; it is the comparison across models that fails.

3. What is a tri-merge credit report?

A combined report mortgage lenders order that draws from all three nationwide bureaus at once, with a score from each. For conventional loans this has long meant the Classic FICO versions, and note those are three different bureau-specific versions, not one model. VantageScore 4.0 became eligible for delivery to Fannie Mae in April 2026 through a limited set of approved lenders, and wider access was still being worked out as of this writing, so ask which model your lender uses.

4. Which of my three mortgage scores is used?

For a single borrower, the convention is the middle score of the three. Mine were 728, 734 and 731, so the file was underwritten on 731. For joint applicants the representative score used in pricing and delivery is commonly the lower of the two borrowers' middle scores, while eligibility can depend on the program and the underwriting system.

5. Why do my three bureau scores differ from each other?

Two reasons at once. The three Classic FICO versions are bureau-specific and not identical models, and the underlying files differ because furnishers choose which bureaus to report to. A large spread is worth investigating by comparing the three reports, since the scores alone cannot identify the cause.

6. Can I ask a lender which credit score they use?

Yes, and you should. It is a specific question with a specific answer, and knowing the bureau lets you check that report. If you have a freeze, ask what the lender needs before lifting it, because a mortgage tri-merge requires all three.

7. Should I worry if I am just above a lender threshold?

Yes. If your app shows a number just above a cutoff, you are near a line measured by an instrument you cannot see, and the actual pull can come in lower for ordinary reasons. Build a buffer rather than applying the week you cross it.