Marilyn vos Savant's IQ: the 228, the 186, and what each one measured

0 Quick Answer

Marilyn vos Savant's 228 is a ratio IQ computed from a 1937 Stanford-Binet administered when she was ten, her 186 is an extrapolated score on Ronald Hoeflin's Mega Test, and neither number is a score on the deviation scale that every modern IQ test uses, which is why Guinness retired the category that listed her in 1990. The encyclopedia record and the Wikipedia article, used here as pointers to the biography and not as sources for any figure, agree on the outline: born in St. Louis in 1946, tested as a child, listed in the Guinness Book of World Records as holding the highest IQ from the mid 1980s until 1989, author of the Ask Marilyn column in Parade from 1986, and the correct party in the Monty Hall dispute of 1990 and 1991 that the New York Times reported on its front page.

The arithmetic of the 228 is the whole story. A ratio IQ is mental age divided by chronological age, multiplied by 100, a formula Lewis Terman set out in The Measurement of Intelligence in 1916. A child of ten who performs at the level the test assigns to a person of 22 years and 10 months has a ratio IQ of 228. The same formula gives absurd results for adults, which is why David Wechsler replaced it with the deviation IQ in The Measurement of Adult Intelligence in 1939, and why no test in use today reports a ratio. A 228 is therefore a real historical number that cannot be placed on a modern scale, and the page on mental age explains the two units side by side.

The 186 has a different origin and a different problem. It was reported on the Mega Test, a take home, untimed high range test whose norms the Mega Society itself describes as extrapolations from reported scores on supervised tests. A number in that region is beyond the ceiling of any normed instrument: the WAIS-5 reports composites up to 160, and ACIS reports to a ceiling set by its norm frame and compresses rarity language above it. The record, taken as a whole, documents a very able person and two numbers that do not measure what a modern IQ measures.

1 Where the 228 Comes From, in Her Own Account

The 228 has a traceable origin, which is rare among famous IQ numbers, and the origin is a ratio computed from a children's test in 1956. By her own account, reported in the biographical sources cited above, her first test was administered in September 1956, when she was ten, using the 1937 revision of the Stanford-Binet. The examiner recorded a mental age of 22 years and 10 months. Under the ratio formula in use for that instrument, mental age divided by chronological age and multiplied by 100 gives 22.83 divided by 10, times 100, or 228. The number is not an error and not a fabrication. It is exactly what the formula produces for a ten year old who answers items assigned to adults.

What the formula produces is not what a modern reader assumes. Terman's 1916 manual explains the logic: a mental age was the age level at which a child's performance matched the typical child in the standardization sample, and dividing by chronological age expressed how far ahead or behind a child was in developmental terms. The quotient was designed to describe children. It says that a ten year old performed like the test's model of a 22 year old, and it says nothing about how rare that performance is among ten year olds, because a ratio is not a rank.

The distinction matters because every later use of the number treats it as a rank. The page on how IQ is calculated explains the modern method: a raw score is compared with the distribution of scores among people of the same age, and the result is placed on a scale with a mean of 100 and a standard deviation of 15. On that scale, a 228 would be more than eight standard deviations above the mean, a point at which the normal distribution predicts fewer than one person in the entire history of the species. The 1956 examiner did not claim that. The ratio formula simply does not map onto the deviation scale at its extremes, and the page on the 15 point standard deviation shows why the two units diverge as scores rise.

The 1937 Stanford-Binet itself has a documented history. Terman and Maud Merrill published it as the second revision of the scale, and its ratio IQs were known at the time to be unstable at the ages and levels where mental age outran the test's adult norms, which is one reason the 1960 revision moved the Stanford-Binet to deviation scoring. The test that produced the 228 was replaced by its own authors within a few years of her sitting it.

2 The 1937 Stanford-Binet, and the Problem Its Authors Knew About

The instrument that produced the 228 was a serious test with a documented standardization, and its own authors and reviewers identified the weakness that makes a very high ratio hard to interpret. Lewis Terman and Maud Merrill published the second revision of the Stanford-Binet in 1937 as two parallel forms, L and M, described in Measuring Intelligence, with a new standardization sample of American children and adults drawn to be more representative than the 1916 sample. Quinn McNemar's 1942 analysis of the standardization data, The Revision of the Stanford-Binet Scale, examined the statistical properties of the new forms in detail, including the variability of IQ at different ages and levels.

The problem is in the scale's construction. A ratio IQ depends on mental age, and mental age is credited by passing items assigned to age levels up to the test's top, which in the 1937 revision extended into superior adult levels. A ten year old who passes items at the superior adult levels is credited with a mental age far above ten, and the ratio multiplies that gap. Because the spread of mental ages is not the same at every chronological age, the same ratio corresponds to different rarities at different ages, and McNemar documented that the standard deviation of ratio IQs varied across the age range of the 1937 sample. A 228 at age ten and a hypothetical 228 at age six would not describe the same rank, and neither can be converted to a rank without the age specific spread, which is exactly what a deviation scale supplies.

The authors responded in the next revision. The 1960 Stanford-Binet, Form L-M, reported deviation IQs computed with a mean of 100 and a standard deviation of 16, abandoning the ratio for the reasons Wechsler had given in 1939 and McNemar's analysis had quantified. The page on types of IQ tests traces the later editions to the Stanford-Binet 5 of 2003, and the page on the history of IQ sets the ratio to deviation change in the wider story of the scale. Every Stanford-Binet score reported since 1960 is a deviation score, which means that the 228 belongs to the last generation of the test that could have produced it.

None of this diminishes the 1956 performance. It locates it. A ten year old passed items written for superior adults on a well built instrument, and the examiner recorded what the manual said to record. The number that resulted is a fact about the test as much as about the child, and reading it requires knowing both.

3 What a Ratio IQ Means, and Why It Was Abandoned

A ratio IQ compares a child's performance to an age level, a deviation IQ compares a person's performance to their own age group, and the second replaced the first because the first breaks for adults and inflates at the top. Wechsler set out the case in 1939 in The Measurement of Adult Intelligence, the book that introduced the scale that became the Wechsler Adult Intelligence Scale. His objection was arithmetic. Mental age stops rising in adulthood, so a ratio IQ falls with every birthday for anyone over the age at which the test's mental age scale tops out, and a forty year old with the same raw performance as a twenty year old would receive half the IQ. His solution was to define IQ as a standard score within an age group: 100 for the median, 15 points for each standard deviation.

The deviation scale fixed adults and changed the meaning of every number. A deviation IQ of 130 means two standard deviations above the median of the age group, which is the 98th percentile, the figure Mensa uses for admission. A ratio IQ of 130 means a mental age 30 percent ahead of chronological age, which has no fixed percentile at all, because the spread of mental ages differs at every age. The two scales agree near 100 and diverge as they rise, and by the region of 200 they describe different quantities. The IQ score chart lays out the deviation scale; there is no equivalent chart for ratios, because a ratio was never a rank.

This is the sense in which the 228 is real and unusable. It documents an exceptional performance by a ten year old on a 1937 instrument. It does not document a position on any scale that a test today would report, and the honest answer to what her modern deviation IQ would have been is that nobody knows, because no such measurement was made at the time and none can be made retrospectively. The page on how IQ scores are normed explains why a score is inseparable from the sample and the scale that produced it.

4 The Mega Test and the 186

The 186 is a different kind of number from the 228: not a ratio but an extrapolation, from a test built by one man to discriminate at the one in a million level, with norms its own society describes as inferred rather than sampled. Ronald Hoeflin published the Mega Test in the mid 1980s as a take home, untimed set of very hard verbal and spatial problems, and used it as the admission instrument for the Mega Society, which describes itself on its own site as an organization of people who have scored at the one in a million level on a test of general intelligence which is credibly claimed by its authors to be able to discriminate at this level. The same site lists the tests it accepts, the Mega Test, the Titan Test, the Ultra Test and the Hoeflin Power Test, and describes their norming as Hoeflin's norming of the Mega and Titan tests extrapolating from reported scores on supervised, timed tests.

That sentence is the whole evaluation. A normed test samples a reference population and converts raw scores to ranks within it, as the page on reliability and validity explains. An extrapolated norm takes the scores that test takers report having obtained on other instruments, relates them to raw scores on the new test, and extends the line beyond the range where any instrument has people to sample. The result can be internally consistent and still describe no population, because no population of one in a million was ever measured. The page on high range IQ tests covers the genre and its limits in detail.

The Mega Test's conditions add a second layer. It was untimed and taken at home, with no supervision and no control over reference materials, collaboration or repeated attempts, which the International Test Commission's guidelines would classify as open mode, the weakest basis for interpreting a score. The Mega Society's own admission page reinforces the point by listing which tests it accepts and describing how they were normed, which is more than most high range societies publish and is exactly what a reader needs in order to read the number. A score from a test normed by extrapolation is an ordinal statement: the person answered more of these problems than most of the people who submitted answers. It is not a percentile of the population, because the population was never sampled, and it is not a deviation IQ, because a deviation IQ is defined by a sampled distribution.

None of this is a criticism of vos Savant, who took the test that existed and scored highly on it. It is a description of what a 186 on that instrument can support: very strong performance on a set of hard problems, under conditions that cannot bound the score, converted to a number by a line drawn past the last data point.

5 What Guinness Listed, and Why It Stopped

Guinness listed her as the holder of the highest IQ from the mid 1980s until 1989 and retired the category in 1990, and the retirement is the most informative fact in the record. The biographical sources agree that the Guinness Book of World Records carried her under the highest IQ heading through those editions and that the publisher then dropped the category, on the stated grounds that IQ tests were too unreliable to designate a single record holder. The category has not returned, and Guinness World Records lists no highest IQ today, which is why every subsequent claim to the title, including the ones reviewed on the page on the highest IQ ever, rests on bodies that are not Guinness.

The publisher's reasoning was correct and applies to both of her numbers. A ratio IQ from 1956 and an extrapolated high range score from the 1980s are not measurements on the same scale as each other or as any modern instrument, and a record book cannot rank quantities that are not commensurable. The same arithmetic that makes a 228 unplaceable makes any highest IQ record unplaceable: above the ceiling of a normed test, there is no sample, and without a sample there is no rank. A record requires a rank.

The print record is also the reason the number is so durable. A Guinness edition sat in millions of homes and school libraries, and a figure printed there for four consecutive years acquires the authority of the book rather than of the test behind it. Readers who met the 228 in Guinness met it without the words ratio, 1937 or age ten, and the internet inherited the number in the same stripped form. The category's retirement did not retire the number, because a retirement is not reprinted the way a record is.

What Guinness could have listed, and never did, was a documented deviation score on a normed instrument with a published ceiling, which for the Wechsler adult scale is 160. Nobody holds a record on that scale either, because a ceiling is a floor for the people above it: everyone who exceeds it receives the same number, and the instrument cannot say who is highest. The page on the gifted IQ range explains what the top of a normed scale can and cannot resolve.

6 A 1956 Norm Read in 2026

Even if the 228 could be placed on a deviation scale, it would have to be placed on the 1937 norms it was scored against, and those norms are nearly nine decades old. James Flynn documented in Psychological Bulletin in 1987, volume 101, issue 2, pages 171 to 191, that raw performance on intelligence tests rose across 14 nations through the twentieth century, so that each generation scored higher against the norms of the previous one. Trahan, Stuebing, Fletcher and Hiscock meta-analysed the effect in the same journal in 2014, volume 140, issue 5, pages 1332 to 1360, and estimated the gain at about 2.3 points per decade. The page on the Flynn effect works through the consequences for any score computed against old norms.

The arithmetic for a 1937 standardization read in 1956 is a drift of about two decades, or roughly 4 to 5 points by Trahan's estimate, which is our arithmetic on the published rate and is small relative to a 228. The arithmetic for a 1937 standardization read today is a drift of nearly nine decades, roughly 20 points on the same estimate. Neither figure changes the conclusion that a ratio cannot be placed on a deviation scale, but both illustrate a second reason a number from 1956 cannot be compared with a number from 2026: the reference population moved. A test today compares a person with contemporaries measured recently, which is why the page on how IQ scores are normed treats the norm year as part of the score.

There is a third reason, which is the test itself. The 1937 Stanford-Binet, the Mega Test and a modern battery measure overlapping but different content with different item formats, and the page on the g factor explains why scores on different instruments correlate highly without being interchangeable. A modern battery would report the domains separately, and the single number that a 1956 examiner wrote down would appear as a profile of verbal, reasoning, quantitative, spatial, memory and speed scores, each with its own band. That profile is the thing the 228 stands in for, and it does not exist.

7 What the Record Documents: the Column and the Monty Hall Dispute

The best documented cognitive act in vos Savant's public life is a probability argument she won against thousands of correspondents, and it is documented because a newspaper of record covered it. From 1986 she wrote the Ask Marilyn column in Parade, the Sunday magazine distributed with hundreds of American newspapers. In 1990 a reader posed the Monty Hall problem: a contestant picks one of three doors, the host, who knows where the prize is, opens another door to reveal a goat, and the contestant is offered the chance to switch. She answered that switching wins two times in three. The response, by her account and by the press coverage, ran to some ten thousand letters, many from readers with advanced degrees, most telling her she was wrong.

She was right, and the argument is short. The contestant's first choice is correct one time in three. The host's action does not change that probability, because the host always has a goat to show. The other unopened door therefore holds the prize two times in three. John Tierney's front page report in the New York Times on July 21, 1991, headlined Behind Monty Hall's Doors: Puzzle, Debate and Answer?, set out the dispute, the simulations that confirmed her, and the mathematicians who conceded. The problem is now a standard teaching example in probability, and the page on the cognitive reflection test discusses why intuitions of exactly this kind mislead most people.

The episode documents a specific ability: reasoning correctly about conditional probability under public pressure and defending the reasoning against credentialed disagreement. It does not document an IQ, and the page on what an IQ test measures explains why a single demonstrated act, however impressive, is not a measurement across domains. It does document that the person listed by Guinness could do the thing that her critics, with their doctorates, could not, which is a better piece of evidence about her than either of her numbers.

8 The Rest of the Record

Beyond the column, the record consists of books, public appearances and a long career as a writer on reasoning, which is a body of work rather than a test score, and the two are different kinds of evidence. She published books on logic, mathematics and writing, including a volume on the Monty Hall problem and its reception, and continued the column for decades. She married Robert Jarvik, the designer of the artificial heart that bears his name, in 1987, and the encyclopedia record cited above covers the biography in detail. None of this is in dispute, and none of it converts to a number.

The page on signs of high intelligence discusses what documented achievement can and cannot indicate about measured ability, and the page on genius IQ discusses why the word genius attaches to people whose measured scores, where they exist, are often unremarkable relative to their work. Vos Savant's case is the unusual reverse: measured scores that circulate more widely than the work, on scales that do not exist any more.

There is also a distinction to draw between two kinds of evidence that her case puts side by side. A test score is a measurement taken on one day under stated conditions, with an error band whether or not anyone reports it. A career is an accumulation of acts over decades, each observed by other people and most of them recorded. The second kind of evidence is harder to fake and harder to summarize, and the page on the importance of IQ explains why psychologists treat measured ability and demonstrated achievement as related but separate quantities. In vos Savant's case the career is the stronger evidence and the score is the more famous.

What a reader can take from the record is a calibrated picture. A child tested in 1956 performed far beyond her age. An adult in the 1980s scored near the top of a hard, unnormed test. A columnist in 1990 reasoned correctly where most people, including many mathematicians, did not. Each of those is documented. The 228 and the 186 are the least informative items on the list, because they are the two that cannot be placed on any scale a reader would recognize.

9 Why the Number Persists

A famous IQ figure survives on repetition, not on evidence, and vos Savant's numbers survive better than most because they had an institutional origin to repeat. The 228 appeared in print in a record book that did not explain its scale. The 186 appeared in the newsletters and lists of high range societies whose norming their own site describes as extrapolation. Each number then passed into lists of the highest IQs in history, where it sits beside estimates for historical figures who were never tested, and the lists cite each other. The page on celebrities with the highest IQ sorts that genre by evidence class, and the page on common myths about IQ tests covers the belief that a single number can be the highest.

The mechanism is visible in a case reviewed elsewhere in this section. In June 2013 a wire agency reported that Mensa International had published a list of celebrity IQs, the report was repeated by major outlets within a day, and Mensa International stated the following day that it had released no such list; the page on Shakira's IQ documents the sequence. Vos Savant's numbers did not need that mechanism, because a record book printed them, but they are sustained by it now: a search returns pages that repeat 228 and 186 without the words ratio or extrapolated, because those words do not fit a headline.

The audit method that this section applies is the corrective. Find the origin of each number. Name the instrument, the date, the examiner and the scale. Ask whether the number could be placed on a modern scale and, if not, say so. Vos Savant's case is the rare one in which every step can be completed, and completing it leaves two real numbers that measure things a modern reader would not expect and a documented career that measures more than either.

10 Every Number on This Page, With Its Label

A page that grades other people's numbers owes the reader a grade on its own, so the table below labels every figure above by what kind of evidence stands behind it. The labels follow the convention used across this section: documented means a primary record exists and is linked; reported means the figure rests on the subject's own account or on secondary sources that agree; extrapolated means the number was produced by extending a scale past its data; and unplaceable means the number cannot be put on any modern scale regardless of its origin.

The last row is the one that the search results omit. No administration of a modern deviation instrument to Marilyn vos Savant has been published, and none is implied by either of her numbers. A reader who wants to know what her WAIS-5 or Stanford-Binet 5 score would be is asking a question that has no answer, and the page on types of IQ tests explains why scores from different instruments and eras cannot be converted into each other.

11 What It Would Take to Place Her on a Modern Scale

The claim that she has the highest IQ is not unfalsifiable; it is untestable, because the only measurement that could support it does not exist and could not be taken now in a way that would settle anything. A modern deviation score would require an administration of a normed instrument to her as an adult, under the instrument's conditions, reported with its standard error. The instrument would have a ceiling, 160 on the WAIS-5, and a person of her documented ability would plausibly reach it, at which point the instrument would report 160 and say nothing further. The page on how the Stanford-Binet 5 works describes the extended scoring some instruments offer above the standard ceiling, and even that scoring reports a range rather than a rank.

The practical consequence is that no test could confirm or refute the title, because the title is defined in terms that no test produces. What a test could do is what it does for everyone: report a profile across domains, with a band around each score, on a scale with a stated mean and standard deviation. That report would be informative about her in the way that a report is informative about anyone, and it would not be a record.

A reader can measure the thing they are actually curious about, which is usually not her score but their own. The ACIS technical manual states the instrument's ceiling, its norm frame of 3,243 adults from 16 to 90, and the reliability and standard error of each score, and the page on how rare a given IQ is converts any reported score to its frequency on the deviation scale. Neither tool can place a ratio IQ from 1956, and neither pretends to.

12 Measuring Your Own Profile Instead of Ranking Hers

The question behind most searches for a famous person's IQ is answerable for the person searching, and the answer is a profile with a band rather than a single number with a title. ACIS is a normed online battery of 20 subtests across six domains, scored against an adult reference frame of 3,243 records in age bands from 16 to 90, and its technical manual, version 1.4 updated August 3, 2026, reports a Full Scale IQ composite reliability of .9886 with a standard error of 1.60 points and a reliability and standard error for every index and subtest. We sell it, which is a disclosure the reader should hold against everything on this page, and it is not accepted for Mensa admission or for any institutional purpose.

Its scale is the modern one. Scores are deviation scores with a mean of 100 and a standard deviation of 15, the ceiling is set by the norm frame, and the manual states that rarity language above the ceiling is compressed rather than extrapolated, which is the opposite of what the Mega Test's norming did. A person who reaches the ceiling receives the ceiling and a note that the instrument cannot resolve further, which is the honest report and the reason no ACIS score will ever be a 228.

The forms are priced one time: 15 dollars for six subtests, 30 for thirteen, 50 for all twenty, and a per subtest price for a custom selection on the order page. Five subtests are free without a card, access to a purchased form stays open for 30 days, and the quality guarantee is a full refund within five days if the report does not deliver the features described at checkout or a technical issue prevents access. The limits are the ones stated across this site: self administered rather than supervised, normed on English speaking adults, and a Visual Spatial Index affected by the retirement of one subtest on September 7, 2026 pending the next edition's tables.

13 Sources Behind This Page

Every figure above is traceable to one of the following, and each is linked at the point where it is used. Wikipedia and the encyclopedia entry are cited as pointers to the biography and not as sources for any IQ figure; the ratio arithmetic on the 228 is ours, applied to the mental and chronological ages she has reported, and is labelled as such.

• Terman L M. The Measurement of Intelligence: An Explanation of and a Complete Guide for the Use of the Stanford Revision and Extension of the Binet-Simon Intelligence Scale. Houghton Mifflin, 1916. archive.org.
• Wechsler D. The Measurement of Adult Intelligence. Williams and Wilkins, 1939. archive.org.
• Terman L M and Merrill M A. Measuring Intelligence: A Guide to the Administration of the New Revised Stanford-Binet Tests of Intelligence. Houghton Mifflin, 1937. archive.org.
• McNemar Q. The Revision of the Stanford-Binet Scale: An Analysis of the Standardization Data. Houghton Mifflin, 1942. archive.org.
• Flynn J R. Massive IQ gains in 14 nations: What IQ tests really measure. Psychological Bulletin, 1987, volume 101, issue 2, pages 171 to 191.
• Trahan L H, Stuebing K K, Fletcher J M and Hiscock M. The Flynn effect: A meta-analysis. Psychological Bulletin, 2014, volume 140, issue 5, pages 1332 to 1360.
• Tierney J. Behind Monty Hall's Doors: Puzzle, Debate and Answer? The New York Times, July 21, 1991.
• The Mega Society. Constitution, admission standards and tests accepted for admission, including the description of Hoeflin's norming as extrapolation. megasociety.org, read September 19, 2026.
• Encyclopedia.com. Marilyn vos Savant, biographical entry. encyclopedia.com, read September 19, 2026.
• Wikipedia. Marilyn vos Savant, used as a pointer to the biography and to the Guinness listing years. wikipedia.org, read September 19, 2026.
• Pearson. Wechsler Adult Intelligence Scale, Fifth Edition (WAIS-5), product page, for the instrument's age range and structure. pearsonassessments.com, read September 19, 2026.
• Mensa International. Getting your IQ tested: FAQs, for the 98th percentile admission rule. mensa.org, read September 19, 2026.
• International Test Commission. International Guidelines on Computer-Based and Internet-Delivered Testing. International Journal of Testing, 2006, volume 6, issue 2, pages 143 to 171.
• ACIS. Technical manual, version 1.4, updated August 3, 2026: norm frame, ceiling and rarity language, reliability tables. acisiq.com/technical-manual.

14 Frequently Asked Questions