Is the MBTI accurate? What the reliability and validity research says

Published October 7, 2026Updated October 25, 2026

The MBTI is more accurate than critics usually say about its four scales, and less accurate than fans claim about its sixteen types. Retest studies show the underlying scores are fairly stable, but people near the middle of a scale often land in a different category the next time.

Somebody at dinner says the MBTI is astrology for people with office jobs. Somebody else says it described her better than her own mother could. Both of them took the same questionnaire.

They are arguing about accuracy without having agreed what the word means, and the research on this instrument has the same problem. It goes back to the 1960s, it is large, and it gives two different answers depending on which question you put to it.

I run a free test built on the same four letters, so I have a stake in this. What follows sticks to sources that are named and can be looked up.

Two questions hiding in one word

A test can be accurate in two separate senses, and psychologists keep them apart on purpose.

The first is reliability: does it give the same answer twice? A bathroom scale that reads 150 pounds on Monday and 163 on Tuesday is unreliable, whatever you actually weigh.

The second is validity: is the answer about the thing it claims to be about? A scale that reads exactly 150 every morning, for everybody who steps on it, is perfectly reliable and tells you nothing.

The MBTI does better on the first question than its critics tend to admit, and worse on the second than its fans tend to admit. Most arguments about it are one person talking about reliability and the other talking about validity.

Reliability: the scales hold, the type wobbles

Start with the four scales themselves, before anything is turned into a letter. In 2002 Robert and Mary Margaret Capraro pooled the reliability figures reported across published MBTI studies, in the journal Educational and Psychological Measurement. They found that the scales generally produced strong internal consistency and strong test-retest estimates, with some variation from one sample to the next.

A 2025 review of the current form of the instrument, led by Bradley Erford in the Journal of Counseling and Development, put numbers on the first of those. Its weighted averages for internal consistency ran from .88 to .92 across the four scales, where 1.0 would mean every question on a scale agreed with every other. Those are respectable figures for any questionnaire. The same review also noticed a gap: in twenty-five years of studies, none had reported a retest figure for the current form beyond the ones printed in the publisher's own manual.

Now turn the scales into letters. David Pittenger, reviewing the retest studies in 1993 in the Journal of Career Planning and Employment, reported that as many as half of the people retested after a few weeks came out as a different type.

Both findings can be true at once, and the reason is arithmetic rather than scandal. A type is four separate calls, and it only counts as the same type if all four land the same way. Suppose, purely as an illustration, that each letter holds nine times out of ten. All four holding together is then nine tenths multiplied by itself four times, which comes to about two times in three. A third of people would change type on a retest while every individual scale looked steady.

So somebody reported as ESTJ in March and ESTP in April has not become a different person. One scale out of four crossed its midpoint. Why your result changes between attempts covers which letter usually moves, and why it happens to the people nearest the middle.

Validity: four real dimensions, no real types

The most careful single check on what the letters measure came in 1989. Robert McCrae and Paul Costa gave the MBTI and a Big Five questionnaire to 267 men and 201 women and published the comparison in the Journal of Personality.

Two results came out of it. They point in opposite directions.

The good news for the instrument is that its four scales are not noise. Each one lined up with a dimension of personality that other researchers had arrived at independently, by different methods. Whatever the questionnaire is doing, it is picking up four genuine ways in which people differ.

The bad news is that the study found no support for dichotomies. If introverts and extraverts were two kinds of people, the scores would pile up in two humps with a valley between them. They do not. None of the four scales showed that shape; each is one continuous spread, and the letter records which side of the center line a person happened to land on.

That finding matters more than any other for reading your own result. The description you were handed is written as though you belong to a category. The measurement under it found a position on a line. MBTI vs the Big Five goes through that comparison one scale at a time.

What the outside assessments said

In 1991 a committee of the National Research Council appraised the MBTI in a report called In the Mind's Eye, which looked at training and career development techniques then in use in the U.S. Army. Its conclusion was blunt: "At this time, there is not sufficient, well-designed research to justify the use of the MBTI in career counseling programs."

The committee added that the popularity of the instrument was not matched by supportive research results. It asked for better studies. It did not declare the question closed.

Pittenger came back to the subject in 2005, in the Consulting Psychology Journal, and landed in nearly the same place. The instrument has a great deal of intuitive appeal, he wrote, and may not yet be able to support the claims its promoters make for it.

Read closely, both reviews are aimed at what people do with the result, and at how much gets inferred from a four-letter formula. Choosing careers and sorting staff run well ahead of the evidence. Neither review is an argument that the questionnaire picks up nothing.

What it has not been shown to do

Three claims go past anything those sources support.

It has not been shown to predict who will be good at a job. The publisher itself says the instrument is not to be used for hiring or selection, a point that the history of Briggs and Myers returns to.

It has not been shown that the sixteen types are sixteen kinds of person. The evidence is for four sliding scales, each cut in half for convenience.

And it is not a clinical instrument. It was never designed to detect or rule out any condition, and a free online questionnaire like the one on this site certainly does not.

How to hold your own result

So is it accurate? As a rough reading of where you sit on four real dimensions, on the day you answered, reasonably so. As a permanent category you belong to, no.

The practical upshot is small. Trust the letters where your score was lopsided. Treat a letter that came out close as undecided. Read the description as a portrait of a typical person on your side of four lines, expect parts of it to miss, and be a little suspicious of the parts that flatter, for the reason given in why every type description seems to fit.

None of this makes the exercise worthless. A shared vocabulary for how people differ is worth having even when its edges are soft. The mistake is treating a soft edge as a wall.

MBTI is a trademark of The Myers-Briggs Company, which has no connection to this site. I run it by myself and I am not a psychologist. The test here is a way of describing yourself, and the sixteen type pages are descriptions to hold yourself up against. That is the whole of the claim.