IQ Metrics
IIF Certified Assessment Start IQ Test
IIF Certified Assessment

Cambridge Face Memory Test and IQ: Why Studies Give 0.01 and 0.26

The Cambridge Face Memory Test has 72 items. A 2014 review linked it to IQ at 0.01 and a 2023 meta-analysis at 0.26, while twin studies found 0.70 against 0.29 and a 61% heritability.

Cambridge Face Memory Test and IQ: Why Studies Give 0.01 and 0.26
Illustration generated for IQ Metrics. No photograph is used. IQ Metrics

What you need to know

  • The Cambridge Face Memory Test (CFMT) has 72 items: six faces learned, each picked from a group of three. Typical adults averaged 58 of 72 in the 2006 paper; chance is 24.
  • Its link to IQ depends on the study: 0.01 in a 2014 review of three earlier studies, and 0.26 for face memory in a 2023 meta-analysis of 23 samples (9,062 people), about 7% shared variation.
  • Twin studies: identical twins correlated 0.70 and fraternal twins 0.29 in 289 pairs (2010), and a 2015 study of 2,000 twins put heritability at 61%, mostly unique to faces.
  • Face recognition is a specific ability. It is positively linked to intelligence but mostly its own thing, so a CFMT score is not an IQ and an IQ cannot predict it closely.

The Cambridge Face Memory Test (CFMT) is a 72-question test of learning and recognising new faces. Studies disagree on its link to IQ. A 2014 review put the average correlation at 0.01, which is essentially none. A 2023 meta-analysis of 23 samples and 9,062 people put face memory at 0.26, about 7% shared variation. Neither says face memory is the same thing as IQ.

The twin numbers also depend on the study. A 2010 study of 289 twin pairs found identical twins correlated 0.70 on the test and fraternal twins 0.29. A 2015 study of 2,000 twins put heritability at 61%, and found most of that genetic influence unique to faces. Below: what the test is, why the IQ figures differ, what the twin data show and what it all means for one general factor of intelligence.

What is the Cambridge Face Memory Test?

The CFMT is a test in which you learn six faces, then pick each one out of a group of three. Duchaine and Nakayama introduced it in 2006. Each target face gets 12 test items: 3 with the view you studied, 5 with new views and 4 with new views plus visual noise. Six faces times 12 items makes 72 questions.

In the 2006 paper, 50 typical adults averaged 58 of 72, or 81% by our arithmetic. Upside-down faces dropped the mean to 42. Eight people with prosopagnosia, a severe difficulty recognising faces, averaged 37, and six of the eight scored outside the normal range. Chance is 24 of 72, one in three.

The faces appear without hair or other cues, so the test isolates the face itself (Wilmer and colleagues, 2010). In that study's web sample of 3,004 people, the mean was 75.6% correct. We do not host the CFMT, and nothing on this page diagnoses face blindness.

The CFMT was built to fix two older tests. Duchaine and Nakayama say the Warrington test includes non-face cues, and the Benton test shows faces side by side, which lets people match single features. In their paper, those two older tests failed to classify most of the people with prosopagnosia as impaired, while the CFMT did.

Is face recognition linked to IQ?

On face recognition and IQ, the answer is a little by the newest count, and not at all by an older one. In 2014, Jeremy Wilmer, Laura Germine and Ken Nakayama reported a mean correlation of 0.01 between the CFMT and standard IQ tests. It was weighted by sample size and corrected for range restriction in the IQ tests, and it pooled three earlier studies.

In 2023, Dana Walker, Romina Palermo, Zoe Callis and Gilles Gignac pooled far more data. They estimated a true-score correlation of 0.26 between intelligence and face memory, from 23 samples and 9,062 people. They found 0.42 for face perception (11 samples, 2,528 people) and 0.20 for face detection (2 samples, 407 people).

Why the gap? The 2023 authors say inconsistent results may come from sample size and from the quality of the intelligence measure. Better IQ measures gave higher correlations. By our arithmetic, r = 0.26 means about 7% shared variation and r = 0.01 means 0.01%.

For scale, the 2014 review reports that the CFMT shares 7% of its variation with recognising abstract art images (4,475 people) and 3% with recognising word pairs (3,003 people). So 0.26 puts IQ about level with abstract art. About 93% of the variation in face memory is something else.

A low link could mean the test is weak. The 2014 authors say it is not. The CFMT is as reliable per unit of time as the most widely used IQ test, and it correlates 0.60 with the Cambridge Face Perception Test (124 people) and 0.52 with the Cambridge Famous Faces Memory Test (1,219 people). It also correlates 0.37 with people's own rating of how well they recognise celebrities (190 people).

At 0.26, face memory and IQ share about 7% of their variation: not none, and not most.

Is face recognition genetic?

Is face recognition genetic? Partly, and the size depends on the study. The 2010 twin study tested 289 pairs aged 18 to 57: 164 identical and 125 same-sex fraternal. CFMT scores correlated 0.70 in identical twins and 0.29 in fraternal twins. A standard twin model attributed all of the family resemblance to additive genetic effects, with a 95% interval of 87 to 100%.

The test is not perfectly reliable. Its internal consistency was 0.89 to 0.90 and its retest correlation after six months was 0.70. The authors read the 0.20 gap between the 0.70 twin correlation and the best reliability figure as a modest effect of non-family environment.

A 2015 study by Shakeshaft and Plomin tested 2,000 UK twins aged 18 and 19. It found a heritability of 61% for face recognition. Most of that genetic influence was unique to faces, not shared with object recognition or general cognitive ability.

Heritable does not mean fixed. It describes how much of the variation in the groups studied lines up with genes, not what any one person can do. Our twin studies and IQ heritability article explains the method.

What does face recognition say about one general factor of intelligence?

That the general factor, g, is a strong summary and not the whole story. In 2014 Wilmer's team used face recognition as their example of a specific ability: one that correlates little with g. They argued that many such abilities could be missed. Read the g factor explainer for the background.

The 2023 meta-analysis takes a middle position. It says face processing may fit within the Cattell-Horn-Carroll model, as a narrow ability that is positively linked to intelligence but also distinct, like reading comprehension. Our CHC theory article shows how that model sits behind modern tests.

The genes matter here too. The 2014 review calls face recognition an exception to the generalist gene idea, that most cognitive variation comes from the same set of genes, because its heritability is high yet specific. The 2015 genetic analysis points the same way. For a related split between abilities, see specific abilities against g and working memory and reasoning.

  • 0.01: mean correlation with IQ tests in the 2014 review (three earlier studies)
  • 0.26: face memory and intelligence in the 2023 meta-analysis (23 samples, 9,062 people)
  • 0.70 and 0.29: identical and fraternal twin correlations on the CFMT (2010, 289 pairs)
  • 61%: heritability of face recognition in 2,000 UK twins (2015)

What can a face memory score not tell you?

It tells you how you did on one test of learning new faces, compared with other people. It does not give an IQ, and an IQ cannot predict it closely. If you often fail to recognise people you know, ask a clinician: prosopagnosia is a specific impairment of the recognition of facial identity.

A percentile means the same on any test: the share of people who score below you. Our IQ percentile calculator shows it on a bell curve. Other specific traits have been checked against IQ too, such as aphantasia, and the multiple intelligences theory page explains why the idea of separate intelligences never entered testing.

Method, 18 November 2026: we read the full text of the 2014 review and the 2010 twin study, the abstract of the 2006 test paper, the 2015 twin paper and the 2023 meta-analysis. We did not read the full text of the 2023 meta-analysis or of the three studies behind the 0.01. The 0.26 is a true-score estimate, so it is higher than a raw correlation. The 81%, the 7% and the 93% are our arithmetic.
Your own number

Where would your own score land?

Face memory is one narrow ability. To see how your overall reasoning compares with other adults, take the IQ Metrics test.

Find your IQ score now! →
Secure & encryptedInstant results10–20 minutes

To train memory in a lighter way, try our memory game, which does not measure face recognition. To place your reasoning on a bell curve, take the IQ Metrics test.

Common questions

What is the Cambridge Face Memory Test?

It is a 72-item test in which you learn six faces and then pick each one from a group of three, in new views and with visual noise. Duchaine and Nakayama introduced it in 2006. Chance is 24 of 72.

Is face recognition linked to IQ?

Somewhat, by the newest count. A 2014 review reported a mean correlation of 0.01, but a 2023 meta-analysis of 23 samples (9,062 people) found 0.26 for face memory, about 7% shared variation.

Is face recognition genetic?

Partly. A 2010 twin study found 0.70 for identical and 0.29 for fraternal twins, and a 2015 study of 2,000 twins put heritability at 61%, mostly unique to faces. Heritable does not mean fixed.

Does a poor face memory score mean a low IQ?

No. Even at a correlation of 0.26, about 93% of the variation in face memory is unrelated to IQ, by our arithmetic. Trouble recognising people you know is a question for a clinician.

Sources for this story

  1. Face recognition: a model specific ability (CFMT and IQ 0.01; Wilmer, Germine, Nakayama) — Frontiers in Human Neuroscience, 2014
  2. Human face recognition ability is specific and highly heritable (289 twin pairs; 0.70 and 0.29) — Proceedings of the National Academy of Sciences, 2010
  3. The Cambridge Face Memory Test: results for neurologically intact individuals and an investigation of its validity (72 items) — Neuropsychologia, 2006
  4. The association between intelligence and face processing abilities: a conceptual and meta-analytic review (r' = 0.26) — Intelligence, 2023
  5. Genetic specificity of face recognition (2,000 twins; 61%) — Proceedings of the National Academy of Sciences, 2015
  6. IQ Metrics computation, 18 November 2026: 12 items per face, 81%, 7% and 93% — IQ Metrics

Corrections: spotted an error? Email corrections@iqmetrics.org and we will update this story and note the change here.

Share this story

Know someone who keeps seeing this number quoted without the scale it was measured on? Send it to them — it takes one tap.

Filed under#genetics and heritability#study quality#scores and scales#confidence intervals

Read the research.
Then find your own number.

Our IIF-certified assessment reports your score with its scale, percentile and confidence range — and a breakdown of the cognitive domains behind it.

Start IQ Test →
Secure & encryptedInstant results10–20 minutes