IQ vs ASVAB: AFQT Is Not IQ

0 The Short Answer

Of every standardized exam people try to convert into an IQ score, the ASVAB is the one that deserves the comparison most. It is a multiple aptitude battery, not a subject exam. Its headline number, the AFQT, is built from four subtests that sample verbal and quantitative ability rather than a school curriculum. It reports a percentile against a nationally representative sample of American youth rather than against the self selected pool of people who happened to sit that test. And for four decades, academic researchers who needed a cognitive ability variable for a large American dataset have reached for the AFQT, because it behaves like one.

None of that makes an AFQT percentile an IQ. It makes the gap between them smaller and more interesting than the gap between IQ and, say, a graduate admissions exam. The differences that remain are specific and worth knowing: the AFQT stops at 99, so it cannot resolve anything above roughly the top one percent. Its reference group was tested in 1997. It reports no domain profile, so it cannot tell you whether your rank came from vocabulary or from arithmetic. And its content leans heavily on knowledge you acquired in school, which is exactly the part of cognitive ability that responds to preparation.

So there is a defensible way to translate an AFQT into IQ units and a much longer list of reasons the translation is a rank statement rather than a measurement. This page does both jobs: it gives you the arithmetic people come looking for, in a table, with the norm assumptions written out, and then explains precisely what that arithmetic does and does not license you to say about yourself.

1 Ten Subtests Across Four Domains

Start with the architecture, because most confusion about the ASVAB comes from people picturing a single test. The official ASVAB program describes a battery of ten subtests on the computerized version, nine on paper, organized into four content domains:

• Verbal. Word Knowledge (WK) and Paragraph Comprehension (PC). Vocabulary recognition and short passage understanding.
• Math. Arithmetic Reasoning (AR), which is word problems, and Mathematics Knowledge (MK), which is school mathematics through basic algebra and geometry.
• Science and technical. General Science (GS), Electronics Information (EI), Auto Information (AI), Shop Information (SI) and Mechanical Comprehension (MC). On the paper form, auto and shop are administered as a single combined subtest, which is why the count drops from ten to nine.
• Spatial. Assembling Objects (AO), a mental rotation and part assembly task, the only subtest in the battery with no knowledge prerequisite at all.

Look at that list next to the structure of a professional cognitive battery and the family resemblance is obvious. Verbal comprehension has two dedicated subtests. Quantitative reasoning has two. Visual spatial processing has one. The science and technical block is a large sample of acquired knowledge, which in CHC terms lands squarely in crystallized ability. What is missing is equally informative: there is no working memory task, no processing speed task, and no inductive fluid reasoning task in the matrix or series completion tradition. Assembling Objects is the closest the battery gets to a content free reasoning measure, and it is spatial rather than abstract.

That inventory explains most of what follows. The ASVAB covers roughly two thirds of the domains a modern battery samples, covers them with real subtests rather than proxies, and leaves the remaining third uncovered. If you want the full map of what a broad battery is trying to capture, the breakdown in what IQ tests actually measure is the reference point this page keeps returning to.

2 The AFQT Uses Four of the Ten

Here is the detail that dissolves half the arguments on the internet: the AFQT is not the ASVAB. It is a composite drawn from four of its subtests, and the other six play no part in it. The official scoring documentation names the four as Arithmetic Reasoning, Mathematics Knowledge, Paragraph Comprehension and Word Knowledge.

The published formula weights them unevenly. The two verbal subtests are first collapsed into a single Verbal Expression score, conventionally written VE, and the composite is then computed as AR + MK + twice VE. Doubling the verbal component means the AFQT is, by construction, an equal weighting of verbal and quantitative ability rather than a simple sum of four parts. Everything mechanical, electronic, automotive, scientific and spatial in the battery is excluded from the number that determines whether you can enlist at all.

This has two consequences that people routinely get backwards. First, a candidate can post an unremarkable AFQT and still be an exceptional fit for a technical specialty, because the composites that govern job assignment draw on the six subtests the AFQT ignores. Second, and more relevant to this page, the AFQT's exclusion of the mechanical and technical block makes it more comparable to an IQ measure, not less. Stripping out auto repair knowledge and electronics leaves a verbal plus quantitative core that looks a great deal like the crystallized and quantitative sections of a standard battery.

The raw scores never reach you in raw form. Individual ASVAB subtests are reported as standard scores with a mean of 50 and a standard deviation of 10, a scale chosen so that a 60 sits one standard deviation above the reference average. The AFQT itself is then converted into a percentile from 1 to 99. Those are three different number systems in one score report, and mixing them up is where most bad conversion math begins.

3 Who You Are Actually Being Compared To

A percentile is meaningless without a denominator, and the ASVAB publishes its denominator with unusual clarity. According to the program's norming documentation, current national norms come from the 1997 Profile of American Youth, known as PAY97. Roughly 6,000 American youths aged 18 to 23 were tested under controlled conditions during the summer and fall of 1997, drawn from a screening of more than 90,000 housing units. The resulting norms went into operational use in 2004. A cohort of 18 to 23 year olds tested in 1997 is a narrow frame for anyone reading a score today, and that gap is what adult-normed IQ testing closes by comparing a result against a defined adult reference group spanning a wide age range.

Read that carefully, because it is the single most underappreciated fact about the AFQT. Your percentile does not describe your standing among people who enlist, and it does not describe your standing among people who took the test the same week you did. It describes where your performance would have fallen in a probability sample of the entire American 18 to 23 year old population, military and civilian, college bound and not. An AFQT of 65 means you outperformed roughly 65 percent of that cohort.

That design choice is exactly what a good IQ norm does, and it is exactly what admissions exams do not do. A graduate admissions percentile compares you to people who chose to apply to graduate school, a group already filtered on the ability being measured. The ASVAB reference group was assembled by knocking on doors. It includes the people who never test.

The caveat rides along with the strength. That sample was tested in 1997, which means a person sitting the ASVAB today is being ranked against a cohort of American teenagers and young adults who finished high school before the smartphone existed. Norm vintage is a real interpretive issue for any instrument, and the honest way to state it is that the AFQT percentile is a well defined rank against a well documented population that is now decades old, not a rank against today's twenty year olds.

4 Why This Is the Closest Case on the Shelf

Comparisons between IQ and other exams usually collapse under one of three problems: the exam measures a taught curriculum, the exam norms against a self selected applicant pool, or the exam produces a single score with no structure behind it. The ASVAB has a partial answer to all three, which is why it belongs in a different category from the admissions tests. The contrast is sharpest against the SAT, which can report one percentile relative to the students who actually sat it and another estimated against a national group, so a reported SAT percentile already carries two possible denominators before anyone attempts a conversion.

The last point is the strongest one, and it is not a matter of opinion. When Ian Deary, Paul Irwing, Geoff Der and Timothy Bates published their 2006 analysis of brother and sister differences in the g factor in the journal Intelligence, the raw material was ASVAB data from the national youth panel. They extracted a general factor from the ASVAB subtests and treated it as g, because the battery's structure supports that extraction. You cannot do that with a curriculum exam, and nobody tries.

There is a general point underneath this. Jeffrey Cucina and Kevin Byle reported in the Journal of Intelligence in 2017 that across mental abilities test batteries, a bifactor model, meaning a strong general factor sitting alongside narrower group factors, fit better than the alternative in more than 90 percent of the comparisons they ran. Any battery that samples several distinct content areas tends to yield exactly that structure, and the ASVAB samples four. The general factor is not something the ASVAB was designed to produce. It falls out of the design anyway, which is the deeper reason this comparison is worth taking seriously.

5 How a Military Screening Test Became Social Science's Ability Variable

The reason the ASVAB shows up in economics and epidemiology papers is a historical accident worth knowing, because it explains why the AFQT carries academic weight that no other non clinical test score carries.

In the summer and fall of 1980, the Department of Defense needed updated ASVAB norms and ran a study called the Profile of American Youth. It piggybacked on an existing survey panel, the National Longitudinal Survey of Youth 1979. The NLSY79 documentation records the scale of the operation: 11,914 respondents completed testing, about 94 percent of the survey sample, at more than 400 sites including hotels, community centers and libraries in the United States and abroad, each paid 50 dollars for their time.

That single decision produced something no researcher could have funded on their own: a nationally representative longitudinal panel in which almost every participant had sat the same lengthy aptitude battery under standardized conditions in the same year. Follow those people for forty years and you can relate a cognitive score measured in adolescence to earnings, health, marriage, occupation and mortality. The AFQT became the cognitive variable in that dataset by default, and then by convention.

The scoring has been revised as researchers found problems with it, which is itself a sign of how seriously it is taken. The documentation describes three generations: an original 1981 percentile, a 1989 revision that fixed inconsistencies by converting to standard scores and double weighting the verbal component, and a 2006 version built from item response theory ability estimates and grouped into three month age bands so results could be compared across the 1979 and 1997 panels. Lynne Steuerle Schofield's 2014 paper in Economics Letters deals specifically with measurement error in the AFQT in this dataset, which tells you how granular the methodological attention has become.

The downstream literature is enormous. David Deming's 2017 paper in the Quarterly Journal of Economics on the growing importance of social skills uses this panel. So does the 2015 study by Christina Wraw, Ian Deary, Catharine Gale and Geoff Der in Intelligence linking cognitive ability in youth to health at age 50. When you read a claim about how measured ability relates to life outcomes in America, there is a substantial chance the underlying number is an AFQT percentile from a military aptitude battery.

6 What the Score Is Actually For

Everything above concerns what the ASVAB happens to be good at. Its official job is narrower and worth stating plainly, because purpose is part of validity and the ASVAB's purpose is not diagnosis. Civilian employers run the same selection logic on a far smaller budget of time, which is why the screeners used in hiring look the way they do: the Wonderlic gives 50 questions in 12 minutes and the CCAT 50 in 15, and neither claims to describe anyone's cognitive architecture either.

The program states three uses: determining enlistment eligibility, assigning applicants to military jobs, and helping students explore careers. The AFQT handles the first. Service specific composites, usually called line scores, handle the second. Those composites are built from different combinations of the ten subtests depending on the branch and the occupational field, which is why the same score report can qualify one candidate for a mechanical specialty and another for a clerical one.

Enlistment thresholds sit on top of the AFQT percentile as a set of category bands. Category I covers percentiles 93 through 99, Category II covers 65 through 92, Category IIIA covers 50 through 64 and IIIB covers 31 through 49. Below that, Category IV is subdivided at 21 to 30, 16 to 20 and 10 to 15, and Category V covers 0 through 9. Applicants in Category V are legally barred from enlisting, and federal law caps Category IV accessions at a small share of annual enlistments. Minimum qualifying percentiles differ by branch and by educational credential.

The instrument arrived at that role gradually. The ASVAB was introduced in 1968, adopted across all the services by 1976, substantially revised in 2002 and renormed in 2004 against the PAY97 reference group. Nearly sixty years of continuous operational use on an applicant flow of that size is a quantity of validation data almost no other test in existence can claim, and it is why the program can publish reliability tables and criterion studies rather than marketing language.

Notice what that system is doing. It is a selection and placement instrument, tuned to sort a large applicant flow into jobs efficiently and to screen out candidates unlikely to complete training. It is not built to describe an individual's cognitive architecture, and it produces none of the outputs a diagnostic assessment produces: no index scores, no confidence intervals around a profile, no comparison of relative strengths across domains. A clinician evaluating a learning difficulty cannot use an AFQT for the same reason a recruiter cannot use a clinical report to fill a quota. The instruments were built for different questions.

7 Free, Proctored and Government Run

Test conditions are usually a footnote. For the ASVAB they matter enough to justify a section, because they are the reason its data is trustworthy in a way that consumer test data is not. Nothing on the cognitive side is organized around a single venue; where an IQ test gets taken ranges from a licensed psychologist's office to a proctored Mensa session to a laptop at home, and only the supervised professional version yields a report a school or a court will act on.

Applicants test at Military Entrance Processing Stations, of which there are 65 across the United States and Puerto Rico, or at satellite Military Entrance Test sites housed in federal buildings, National Guard armories and Reserve centers. Identification is required, arriving late means being turned away, and the official guidance is explicit that a recruiter may drive a candidate to the session but is not permitted inside the testing room. Calculators are prohibited, on the stated logic that the battery is an employment test and needs to observe whether a candidate can apply arithmetic principles unaided.

The computerized version is adaptive: item difficulty adjusts to the examinee's demonstrated ability level, which is why a proctored CAT session runs about two hours while covering the same ground a longer paper form would need. There is also an unproctored preview version, the PiCAT, which takes two to three hours and must be confirmed by a proctored verification session before it counts. Scores may be used for enlistment for up to two years from the test date. Retesting is permitted after a one month wait, again after a second one month wait, and after six months for any further attempt.

On the student side, the ASVAB Career Exploration Program is provided free of charge to participating schools and to high school and post secondary students who sit it there. That combination, no fee to the examinee, professional proctoring, standardized conditions, adaptive administration and a documented national norm, is not something you find outside government and clinical testing. It is the reason an AFQT percentile is a genuine measurement rather than an estimate, whatever its limits as a description of intelligence.

8 Where the Measurement Is Strong and Where It Thins

The program publishes its own reliability figures, which is more transparency than most commercial tests offer, and the numbers reward a close look because they are not uniform. Publishing those figures at all is evidence of quality, because reliability sits with validity and norms in what makes an IQ score accurate rather than merely plausible.

The AFQT composite reaches .97 on the computerized form and .94 on paper. Those are high values by any standard, comparable to the full scale composites that professional batteries report, and they are high for a reason: composites average out the noise in their components. Among the four contributing subtests on the adaptive version, Mathematics Knowledge reaches .93, Word Knowledge .92 and Arithmetic Reasoning .91.

Then look at the same subtests on the paper form. Paragraph Comprehension drops to .74. Electronics Information, which is not in the AFQT, drops to .71. A reliability of .74 means a meaningful share of the variance in that subtest score is measurement noise, and it puts a hard ceiling on how confidently anyone can interpret a single paper administered subtest in isolation. The adaptive version recovers most of that gap, lifting Paragraph Comprehension to .86 and Electronics Information to .86, which is a clean demonstration of what adaptive item selection buys you.

Two practical conclusions follow. Interpret the AFQT composite with confidence; it is measured about as precisely as anything in applied testing. Interpret individual subtest standard scores with much less, especially from a paper administration, and never treat a gap of a few points between two subtests as a profile finding. This is the same discipline that applies to any battery, including how a serious cognitive test should report uncertainty around every index rather than around the total alone.

9 The Conversion Everyone Searches For

You came here for a number, so here is the honest version of one. The AFQT is a percentile against a defined national reference population. IQ is a standard score with a mean of 100 and a standard deviation of 15, also defined against a reference population. Both express rank in a roughly normal distribution, so a percentile can be translated into IQ units by pure arithmetic: find the z score that corresponds to the percentile, multiply by 15, add 100.

This table is arithmetic, not psychometrics, and the distinction is the whole point. It answers one question exactly: if your AFQT percentile were reproduced as a rank on a normally distributed scale with mean 100 and standard deviation 15, what number would appear? It answers a different question not at all: what would you score if a psychologist administered a full battery to you next week?

Note what happens at the top of the table. Because the AFQT stops at 99, everything from roughly the 99th percentile upward collapses into one value. An AFQT of 99 is consistent with an IQ of 135 and equally consistent with an IQ of 150. The scale simply has no resolution up there, which matters because 135 corresponds to roughly the top one percent while the rarities further out grow enormous: 145 is about 1 in 740, and 160 is about 1 in 31,000. If you want that ladder laid out properly, the IQ percentile chart does it row by row, and what counts as a good IQ puts the bands in context.

10 Six Reasons That Rank Is Not a Score

The translation above is defensible as far as it goes. Here is exactly how far that is, stated as six specific objections rather than a vague warning about differences.

• The content leans crystallized. This is the substantive objection, and it has a citation. In their 2000 analysis in Learning and Individual Differences, Richard Roberts, Ginger Nelson Goff, Fadi Anjoul, Patrick Kyllonen, Gerry Pallier and Lazar Stankov examined the ASVAB's factor structure and argued that it is dominated by acquired, culturally transmitted knowledge rather than by fluid reasoning. Look at the subtest list again and the argument is hard to dismiss: vocabulary, school mathematics, general science, electronics, auto mechanics. A person who missed algebra will score lower regardless of reasoning capacity. The fluid and crystallized distinction is not academic hair splitting here; it is the main reason two people with identical reasoning ability can post AFQT scores twenty points apart.
• Three domains are simply absent. There is no working memory subtest, no processing speed subtest and no content free inductive reasoning subtest. A full battery samples those separately because they contribute independent information. An AFQT cannot be an IQ estimate for the same reason a photograph of a house is not a floor plan.
• The ceiling truncates the interesting tail. Ninety nine is the last number on the scale. Anyone whose ability sits meaningfully above the top one percent of the 1997 youth cohort receives the same percentile as everyone else up there.
• The reference cohort is fixed in 1997 and aged 18 to 23. Modern batteries norm by age band, so a 45 year old is compared to other 45 year olds. The AFQT compares everyone, at any age, to young adults tested in one specific year. For an 18 year old applicant that is a close match. For a 35 year old career changer it is a different comparison than an age referenced IQ would make.
• It produces no profile. The AFQT is one number covering verbal and quantitative performance combined. It cannot tell you that your verbal comprehension sits well above your quantitative reasoning, which is often the most useful thing a cognitive report says.
• It responds to preparation in ways the underlying construct does not. Reviewing algebra and vocabulary lists genuinely raises Mathematics Knowledge and Word Knowledge scores, and therefore the AFQT. Something real improved when that happens, but it was the measured knowledge, and treating the gain as a change in general cognitive ability confuses the instrument with the thing it samples.

Taken together these do not demote the ASVAB. They locate it: an excellent, precisely measured, population referenced sample of verbal and quantitative developed ability, which is a large and important slice of cognitive ability and not the whole of it.

11 Side by Side With a Full Battery

Laying the two instruments against each other on the dimensions that actually differ makes the relationship easier to hold in mind than any verbal summary. Full batteries also differ from one another on those same dimensions, and a ranked comparison of the top online IQ tests separates them on norm transparency, breadth, score range and the technical evidence each one publishes.

Read down the middle column and the pattern is consistent. The ASVAB is stronger than a cognitive battery on sampling, cost and administrative rigor, and weaker on construct coverage, resolution at the top and diagnostic output. Those are the trade offs of an instrument built to process hundreds of thousands of applicants a year rather than to characterize one person in depth.

12 Reading Your Own AFQT Without Overclaiming

Suppose you have a score report in front of you. Here is what you can legitimately say, in descending order of confidence.

You can say your rank with precision. An AFQT of 78 means your performance on those four subtests placed you above roughly 78 percent of the PAY97 reference cohort, and the composite reliability behind that statement is high enough that a retest would very likely land close to it. That is a stronger claim than most people can make about any cognitive number they hold. The most famous AFQT result in the country shows how fast that rank gets mistranslated, because the roughly 16th percentile Muhammad Ali recorded in 1964 is still repeated as an IQ of 78.

You can say something about the mixture. Because the composite doubles the verbal component, a high AFQT requires reasonable performance on both halves. It is difficult to reach Category I on quantitative strength alone. If your report also shows subtest standard scores, remember that the scale runs mean 50 and standard deviation 10, so a 65 is one and a half standard deviations up, and resist reading a three point difference between two subtests as a strength or a weakness.

What you cannot say is that the number is your IQ, or that it establishes where you sit relative to the general adult population today, or that it describes your working memory, processing speed or fluid reasoning, none of which the battery sampled. If your AFQT surprised you in either direction, the most likely explanations are content coverage and preparation rather than a hidden truth about your intelligence. Comparing the result against what average actually looks like on the IQ scale is a useful sanity check before drawing conclusions, and if your interest is what test scores predict about work rather than what they say about you, the job performance evidence is a better place to look than a conversion table. Being surprised is itself weak evidence, given that self estimates of ability correlate only about .33 with measured scores, which leaves ample room for an honest rank to sit well away from where you expected it.

If what you want is the thing the AFQT structurally cannot give you, a profile across separate cognitive domains with intervals around each one rather than a single rank, that requires a battery built to report it. ACIS is an online self assessment rather than a clinical or official instrument, and it reports results as a domain profile with intervals instead of one bare number.

13 Frequently Asked Questions

14 Sources Behind This Page

Comparing tests means comparing what each one measures. The sources below cover the official test documentation and the research on how admissions scores track cognitive ability.

• U.S. Department of Defense. The official ASVAB site, primary documentation for the AFQT score used in enlistment.
• Frey, M.C. & Detterman, D.K. (2004). Scholastic Assessment or g? Psychological Science, 15(6). The study linking SAT scores to measured cognitive ability, and the reason SAT conversions are estimates rather than measurements.
• Schmidt, F.L. & Hunter, J.E. (1998). The validity and utility of selection methods in personnel psychology. Psychological Bulletin, 124(2). The meta-analysis behind general ability as the strongest single predictor of job performance.
• Koenig, K.A., Frey, M.C. & Detterman, D.K. (2008). ACT and general cognitive ability. Intelligence, 36(2). The companion study on how ACT scores track g.
• Kuncel, N.R., Hezlett, S.A. & Ones, D.S. (2004). Academic performance, career potential, creativity, and job performance: can one construct predict them all? Journal of Personality and Social Psychology, 86(1).
• Study of Mathematically Precocious Youth, Vanderbilt University. A 50 year longitudinal study of five cohorts totalling over 5,000 individuals, identified between the top 3% and the top 0.01% of ability and followed for decades.
• Voncken, L., Albers, C.J. & Timmerman, M.E. (2019). Improving confidence intervals for normed test scores. Behavior Research Methods. Open access. Documents the mean 100, SD 15 metric and the uncertainty that norming from samples adds to any score.
• Crawford, J.R., Garthwaite, P.H. & Slick, D.J. (2009). On percentile norms in neuropsychology. The Clinical Neuropsychologist, 23(7), 1173-1195. Three definitions of a percentile coexist in practice and can return different results for the same score.
• Pearson Clinical Assessment Scientific Council (2023). Standardized Clinical Assessment for Practitioners: A Primer. How standard scores, percentile ranks and the standard error of measurement are meant to be read together.
• Pearson (2008). WAIS-IV Score Report sample. What a real report contains: every composite paired with a percentile rank, a 95% confidence interval and a qualitative description, never a bare number.