The computer ASVAB chooses every question from a running estimate of your ability. Each subtest opens with an item of average difficulty, usually goes harder after a right answer and easier after a wrong one, and reports a score on a common scale no matter which questions you saw. That design is why you cannot skip or go back, and why unfinished items cost you points.
The CAT-ASVAB is built on item response theory, usually shortened to IRT. Each new question written for Forms 5 to 9 was pretested on at least 1,200 examinees and given three numbers: how hard it is, how sharply it separates stronger from weaker test takers, and how likely a very low-scoring person is to get it right by guessing. Technical Bulletin No. 3 calls this the three-parameter logistic model. Your ability is a number on the same scale as the item difficulties, which is what lets the test pick questions that fit you.
Within each subtest the software runs one loop. It picks the item that tells it the most about a person at your current estimate, you answer, and it recomputes the estimate. A correct answer usually moves you up to a harder item and a miss moves you down.
Two subtests follow a fixed content order, according to Technical Bulletin No. 3. General Science alternates physical science and life science items for 15 items. Assembling Objects gives 7 connection items and then 8 puzzle items. The other subtests rely on the mix of content in the pool.
Because every item is matched to you, the computer test needs far fewer of them: 135 scored items across 10 subtests, against 225 on paper.
The next question does not exist until you answer the current one. It is chosen from your updated estimate, and a skipped item gives the software no answer to update from. The same logic rules out going back: every later item was picked because of the answer you gave. The official recruiter guidance says applicants cannot change an answer because the test is adaptive.
So the official prep page tells you to confirm the answer you have selected before you submit it. Instructions work differently. You can review them as often as you need, and time spent reading them does not count against the subtest limit.
Early in a subtest the software knows very little about you, so each answer moves the estimate a long way. Miss the first item and the second will usually be easier; you then need correct answers to climb back. Later the estimate has settled and each answer nudges it less. When the program office pretested new items for Forms 5 to 9, it seeded them in the second to fourth positions of most subtests, which Technical Bulletin No. 3 describes as the stretch before the ability estimate stabilizes.
That early swing is the source of the popular claim that the first five questions decide your score. The claim is wrong. Your final score is a fresh estimate computed from every response in the subtest. The test keeps serving items at your current estimate, so a strong test taker who stumbles early gets harder items again after a few right answers. Late items are where the test measures you most precisely, because by then they are chosen from a settled estimate and sit close to your level.
Tryout questions are new items being tested for future forms. They do not affect your score. According to the official site, each examinee receives tryouts in 2 to 4 of the 10 subtests, and a subtest that carries them gets a longer time limit. In the current table each subtest can carry up to 15 tryout items.
The official site says tryouts are randomly dispersed throughout each selected subtest, and it describes no on-screen label for them. Technical Bulletin No. 3 explains that the pretest seeding positions were chosen to make it hard for examinees to tell when they were seeing a tryout, so that they would give the same effort on every item. A subtest that feels long may be carrying tryouts. Treat every question as if it counts.
Each subtest has its own clock. The time and number of questions remaining appear in the lower right-hand corner of the screen, and when you finish a subtest early you move straight to the next one.
The program office publishes two different averages for the whole test. The What to Expect page says the proctored test takes about 2 hours on average, while the ASVAB Fact Sheet says the average examinee takes about 1 1/2 hours. The limits below are the current official figures; the last column divides the regular limit by the scored items to show your average budget per question.
If the clock expires before you finish, the subtest ends and a penalty procedure is applied. The official site describes it this way: the items you did not complete are scored as though they were answered at random. Technical Bulletin No. 3 adds three properties. The size of the penalty is related to the number of unfinished items. Two people who answered the same number of items and had the same estimate get the same penalty. And the procedure was designed to remove any coachable strategy for artificially raising a score.
The ASVAB subtests are power tests, meant to measure your best work rather than your speed, and the program office sets the limits so that virtually everyone can finish. The official site says that in almost all cases the penalty never has to be applied.
The paper test has no penalty for guessing, so the official advice there is to fill in every blank before time is called. The ASVAB format guide lays out the two versions side by side.
Every testing station has a red HELP key. Press it when you need something, such as more scratch paper or a pencil. While HELP is active the subtest clock stops, and it starts again only when you return to the questions, so time spent in HELP does not count against the subtest limit.
If a computer fails, your test data is not lost. You restart on another computer at the beginning of the subtest that was interrupted, and that subtest's timing starts over. The ASVAB test day guide covers check-in, the testing room and what to bring.
Two applicants can see completely different questions and still earn comparable scores, because IRT places items and people on one scale. The program office does not rely on that alone. When CAT-ASVAB Forms 5 to 8 went into operation on August 25, 2008, their scores had been equated to the reference pool, Form 4, in a three-phase study of more than 100,000 applicants. Linear equating forces every pool to produce the same mean and spread of scores, so the pool you happen to draw does not change what your score means.
The paper forms are tied to the same scale. The 1997 norming study was run on a computer form, CAT-ASVAB Form 04D. Technical Bulletin No. 4, section 5.2, describes a direct linear equating of paper Form 25B to Form 04D, with the resulting transformations applied to the other paper forms in the 23 to 27 series. Each subtest is reported as a standard score with a mean of 50 and a standard deviation of 10.
The official FAQ says the versions differ in difficulty, but the reported scores are statistically linked, so you should expect a similar score on either one. The computer test may feel harder if you are above average, because it keeps serving items near your limit.
Your AFQT comes from four subtests: Arithmetic Reasoning, Mathematics Knowledge, and Word Knowledge with Paragraph Comprehension combined into a verbal score that counts twice. That composite is then converted to a percentile against a fixed reference group, the Profile of American Youth 1997, known as PAY97. It is a nationally representative sample of about 6,000 young people aged 18 to 23 who took CAT-ASVAB Form 04D between June 1997 and April 1998, most of them in the summer and fall of 1997. The norms took effect in July 2004.
An AFQT of 62 means you scored as well as or better than 62 percent of that sample. It is not the percent of questions you answered correctly, and it does not rank you against the people who tested this year. The scale stops at 1 and 99 by rule: in Segall's conversion table, percentiles that round to 0 are reported as 1 and those that round to 100 are reported as 99. The AFQT calculation guide walks through the formula, and the ASVAB score range page covers what each part of the score report can show.
None of this rewards tricks. The penalty function was written to stop coachable shortcuts, and the adaptive design keeps testing you at your level. Each mechanic calls for one habit.
Manage the clock subtest by subtest. Arithmetic Reasoning gives you more than 3 minutes per scored item, so a slow, careful setup pays off. Word Knowledge and Shop Information give you about 36 seconds, so a word you do not know is worth a quick elimination and a choice, not a long stare. Glance at the counter in the corner once or twice per subtest and compare items left against minutes left.
Raising the score itself comes down to the four AFQT subtests. The guide to improving your AFQT score covers how to spend your study time. If you are taking the unproctored PiCAT instead, note that it has no subtest time limits and no tryout items; the PiCAT and ASVAB comparison explains the verification test that follows it.