What is a good AP Statistics score?

By Jude Wallis · Published

It depends what you need the score for. On the last results recorded here, May 2025, a 3 or better was the majority result: 60.3% of students reached it and the mean score was 2.92. Those figures describe the old nine-unit course. The first exam on the new one is May 2027.

This page is about exam scoring rather than a course topic, but the reasoning it uses is Unit 1 material: the mean and median as measures of center, and the limits on what a pair of summary statistics can tell you about a distribution. The May 2025 results quoted here are the most recent recorded on this site and describe the previous nine-unit course; the first exam on the Fall 2026 course is May 2027. The fully digital Bluebook delivery and the content removals are stated on College Board's AP Statistics future revisions page.

The short answer, in three parts

"Good" is three different questions wearing one coat, and they do not have the same answer.

Good compared with other test takers. The most recent AP Statistics results recorded on this site are from May 2025: 60.3% of students earned a 3 or higher, and the mean score was 2.92. So a 3 or better was the majority result, and roughly two in five finished below a 3. The first figure alone puts the median at 3 or above, and the pair of them narrow it to a 3 or a 4; which one it was depends on how many students landed exactly on 3, and that number was never published.

Good enough for college credit. Individual colleges set their own credit thresholds, and they do not agree with each other. This page will not tell you what any particular college accepts, because that is the college's own published policy, it varies by institution, and it changes. Check it at the schools you care about.

Good relative to the effort you can spend. This is the only one of the three you control, and it is the one worth planning around.

Two hard limits sit behind all of this. May 2025 was an exam on the nine-unit course that the redesign replaced, so no score distribution exists for the exam most readers of this page will actually sit, which is May 2027. And no table converting a raw point total into a 1 to 5 score has been published for the redesigned course. Any specific threshold you are shown for it was produced by guesswork, however confidently it is presented.

The last recorded results, and which exam they describe

FigureMay 2025
Share earning 3 or higher60.3%
Mean score2.92
Share earning below 339.7%

The third row is just the first subtracted from 100%. Everything else on this page is built from the first two.

Those results belong to the previous course. AP Statistics was rebuilt effective Fall 2026: nine units became five, covering 55 topics, five blocks of content were removed, and College Board's future revisions page states that the May 2027 exam moves from hybrid digital to fully digital, with all student responses submitted through the Bluebook testing app. What changed in AP Statistics for 2027 works through the whole list.

So read 60.3% and 2.92 as a baseline for the exam that came before, not as a forecast. A different syllabus, sat by a different cohort in a different format, does not have to land in the same place, and nobody gains from a confident guess about which way it moves. What the old figures are good for is calibration: they tell you what kind of score distribution this subject has produced, which is more than you can say for a number invented for 2027.

What a mean of 2.92 and a 60.3% share actually pin down

Two summary numbers cannot recover a five-category distribution. There are five unknown shares and only three equations available, counting the requirement that the shares add to 1. But those two numbers are not empty, and it is worth being precise about what they force.

The median score was a 3 or a 4. The share at or below 2 was 39.7%, which is less than half, so the median is at least 3. It does not go further than that. The 60.3% figure counts students at or above 3, not at or below it, so the share at or below 3 is 39.7% plus however many earned exactly a 3, and that last number was never published. If at least 10.3% landed exactly on a 3, the median is a 3; if fewer did, it is a 4. Cohort B in the next section has nobody at a 3, matches both published figures, and has a median of 4. What the pair does settle is that the median is not a 1, not a 2 and not a 5, and the mean is needed only to rule out the 5.

The mean sat 0.08 below a score of 3. Write it as an accounting identity. If p1p_1 through p5p_5 are the shares earning each score, then the distance of the mean from 3 is (p4+2p5)−(2p1+p2)=2.92−3=−0.08(p_4 + 2p_5) - (2p_1 + p_2) = 2.92 - 3 = -0.08. In words, the 1s and 2s pulled the mean below 3 slightly harder than the 4s and 5s pulled it above. That 0.08 is the gap from a score of 3, not from the median. The mean did sit below the median either way, but the gap was 0.08 only in the branch where the median was a 3; in the other branch it was 1.08. A mean below the median is a common signal of a longer left tail, not a theorem about one.

At least 15.85% earned a 4 or a 5. This one needs both figures, and the derivation is in the first worked example below. The short version: if every student below 3 had scored a 2 and every student at 3 or above had scored exactly a 3, the mean would have been 2.603. The real mean was 0.317 higher, and only 4s and 5s can supply that surplus. Read it carefully: it is a floor on the 4s and 5s counted together, and it says nothing about how that group split between the two scores.

At most 44.45% earned exactly a 3. That is the 60.3% who reached 3, minus the 15.85% who must have gone past it.

One caveat on the last two. Both published figures are rounded, so treat the bounds as approximate. Even taking the mean as low as 2.915 and the 3-or-higher share at the far end of its rounding, the share earning a 4 or 5 still comes out above 15.5% and the share earning exactly a 3 still comes out below 44.8%.

Put plainly: on that exam a 3 or better was the majority result, and a 4 or 5 put you in a group that could not have been larger than 60.3% of test takers nor smaller than 15.85% of them.

What they do not pin down

Everything else. Here are four cohorts, with each column giving the percentage of test takers at each score. All four have a mean of exactly 2.92 and exactly 60.3% at 3 or higher.

ScoreCohort ACohort BCohort CCohort D
1028.616.039.7
239.711.123.70
344.45027.624.6
4060.317.70
515.85015.035.7

A, B and D are chosen to sit as far apart as the two figures allow, and between them every one of the five shares reaches both its smallest and its largest possible value. A has nobody at a 1 and the largest possible block of 3s. B has nobody at a 3 or a 5, the largest possible block of 4s, and a median of 4. D has the largest possible block of 1s and the largest possible block of 5s. C is an unremarkable shape in between. So the share earning a 5 runs from 0% to 35.7%, the share earning exactly a 3 from 0% to 44.45%, and the share earning a 4 from 0% to 60.3%. Do not read the 15.85% from the previous section as a ceiling on 5s: it is a floor on 4s and 5s together, a different group measured in the opposite direction. The second worked example builds all four.

So when a page shows you a full 1-through-5 breakdown described as the distribution behind a mean and a 3-or-higher share, it has quietly added assumptions you cannot see. This site records the two figures and stops there, because the shape between them is not recoverable from them.

Nobody can tell you the raw score you need

There is no public conversion from a raw point total to a 1 to 5 score for this exam. Composite cut points are not published for the redesigned course, so a sentence like "you need 68% for a 4" is not a fact you are missing, it is a number someone wrote down.

That is worth saying bluntly, because a made-up threshold is worse than no threshold. It sounds checkable, and it will change how you spend your last month: a student who believes they need 68% behaves differently in the final free-response question from one who is simply trying to earn every point available.

What is published is how the exam is weighted, and that is enough to plan against.

SectionQuestionsTimeShare of score
Section I: Multiple-choice4290 min50%
Section II: Free-response490 min50%

Each free-response question is worth 10 points and 12.5% of the exam, so the four of them carry 40 points and half your score. What is on the AP Statistics exam has the per-point arithmetic and the unit weightings, and how to pace the AP Statistics exam turns the timings into a plan. Planning against published weights beats planning against an invented curve, and it has the advantage of being true.

What the 2027 redesign changes here, and what it does not

Changed: the exam those numbers describe no longer exists. The content moved to five units and 55 topics, five blocks were removed, and delivery becomes fully digital in Bluebook. Any score calculator, curve, or distribution you find online was fitted to the old exam. Is the AP Statistics exam digital? is careful about the same distinction, and how many topics are in AP Statistics has the current count.

Changed: which content can even appear. Time spent on removed material is time that cannot earn a point. Two of the five removals were inference procedures, the chi-square goodness of fit test and inference for the slope of a regression line, but inference itself stayed central: Unit 3 is inference for proportions, Unit 4 is inference for means, and question 3 of the free-response section is designated the inference question. See is inference still on the AP Statistics exam and what was removed from AP Statistics, which also explains why the removal list's topic numbers are old numbers that collide with live topics.

Not changed: the cut points are not public. They have not been published for the redesigned course, and a redesign does not make an unpublished number available. Whatever was or was not released for the exam this one replaces, it would not be a guide to a course nobody has sat yet.

Not changed: who decides whether your score is good enough. That is the college, and nothing in the redesign touches it.

Not changed: what a strong performance is made of. Half the score is four written responses, graded on what you actually wrote rather than on what your calculator displayed. That is the part of the exam a study plan can attack, and it does not depend on knowing a curve.

Good enough for what: set the target yourself

The useful version of this question is not "what is a good score" but "what score does the decision in front of me require", and that has an actual answer you can go and get.

  1. List the colleges you are applying to. Look up each one's own AP credit policy. That is the only source that settles what a 3, a 4, or a 5 buys there, and it is published by the college, not by College Board and not by this site.
  2. Note the highest threshold on your list. That is your target. If it is above where practice work currently puts you, you know how much ground to make up and how long you have.
  3. If you have no list yet, aim high and spend the time where the weights are. Half the exam is free response, and the four questions are worth 10 points each, so writing practice is not optional revision.

How to study for AP Statistics turns the unit weightings into a plan, the FRQ guide shows what full credit looks like point by point, and the cram sheet carries the interpretation templates that the written half keeps asking for. If you are still deciding whether to take the course at all, is AP Statistics hard? is the honest version of that question.

What the two published figures force to be true

In May 2025, 60.3% of AP Statistics students earned a 3 or higher and the mean score was 2.92, on a scale of whole numbers 1 through 5. Assuming nothing else, find everything the two figures force about the median score, and the smallest share of students who could have earned a 4 or a 5.

  1. Name the shares. Let p1p_1 through p5p_5 be the proportions earning each score, so p3+p4+p5=0.603p_3 + p_4 + p_5 = 0.603 and p1+p2=1−0.603=0.397p_1 + p_2 = 1 - 0.603 = 0.397.

  2. Median: the cumulative share at or below 2 is 0.397<0.50.397 < 0.5, so the median is at least 3. That is as far as it can be pinned. The 0.603 counts students at or above 3, so the cumulative share at or below 3 is 0.397+p30.397 + p_3, not 0.6030.603. The median is a 3 when p3≥0.103p_3 \ge 0.103 and a 4 when p3<0.103p_3 < 0.103, and p3p_3 was never published. A median of 5 would need p5>0.5p_5 > 0.5, which the surplus accounting below rules out: it caps p5p_5 at 0.3570.357. So the mean is needed only to eliminate the 5.

  3. Now build a floor for the mean. Give every student below 3 a 2, the highest score they can have, and every student at 3 or above exactly a 3, the lowest score they can have.

  4. Floor mean: 2(0.397)+3(0.603)=0.794+1.809=2.6032(0.397) + 3(0.603) = 0.794 + 1.809 = 2.603. Equivalently 3−0.3973 - 0.397, since each below-3 student sits exactly 1 point under 3.

  5. Compare with the real mean: 2.92−2.603=0.3172.92 - 2.603 = 0.317. The cohort's mean was 0.317 above the floor.

  6. Account for that surplus. Against the floor, each student at 4 adds 1, each at 5 adds 2, and each at 1 rather than 2 subtracts 1, so p4+2p5−p1=0.317p_4 + 2p_5 - p_1 = 0.317.

  7. Since p1≥0p_1 \ge 0, it follows that p4+2p5≥0.317p_4 + 2p_5 \ge 0.317. And p4+2p5≤2(p4+p5)p_4 + 2p_5 \le 2(p_4 + p_5), so p4+p5≥0.3172=0.1585p_4 + p_5 \ge \frac{0.317}{2} = 0.1585. Read the same identity the other way for a ceiling on the 5s: 2p5=0.317+p1−p4≤0.317+0.397=0.7142p_5 = 0.317 + p_1 - p_4 \le 0.317 + 0.397 = 0.714, so p5≤0.357p_5 \le 0.357, which is what closed off a median of 5 in step 2.

  8. Check the bound is reachable. Take p1=0p_1 = 0, p2=0.397p_2 = 0.397, p3=0.4445p_3 = 0.4445, p4=0p_4 = 0, p5=0.1585p_5 = 0.1585. Its mean is 2(0.397)+3(0.4445)+5(0.1585)=0.794+1.3335+0.7925=2.922(0.397) + 3(0.4445) + 5(0.1585) = 0.794 + 1.3335 + 0.7925 = 2.92, and 0.4445+0+0.1585=0.6030.4445 + 0 + 0.1585 = 0.603 scored 3 or higher. Both figures match, so 15.85% cannot be improved on without more information.

  9. Allow for rounding. Both published figures are rounded to the digits shown. Running the same argument at the least favorable ends of those rounding ranges still gives a share above 15.5%, so the conclusion survives.

The median score was a 3 or a 4: the two figures rule out 1, 2 and 5, but they do not separate 3 from 4, and that turns on whether at least 10.3% earned exactly a 3. At least 15.85% of students earned a 4 or a 5, and that bound needs both figures. Allowing for rounding, at least about 15.5%. Neither result can be sharpened without the full distribution.

Four cohorts that fit both figures and look nothing alike

Build score distributions over 1 to 5 that all have a mean of 2.92 and 60.3% at 3 or higher, and make them as different from each other as those constraints allow.

  1. Reuse the three relations from the previous example: p1+p2=0.397p_1 + p_2 = 0.397, p3+p4+p5=0.603p_3 + p_4 + p_5 = 0.603, and p4+2p5−p1=0.317p_4 + 2p_5 - p_1 = 0.317. Three equations, five unknowns, so two choices are free. Pick p1p_1 and p5p_5, then p4=0.317+p1−2p5p_4 = 0.317 + p_1 - 2p_5, p3=0.603−p4−p5p_3 = 0.603 - p_4 - p_5, and p2=0.397−p1p_2 = 0.397 - p_1. Free only inside the region that keeps every derived share at or above zero, though: 0≤p1≤0.3970 \le p_1 \le 0.397, p4=0.317+p1−2p5≥0p_4 = 0.317 + p_1 - 2p_5 \ge 0, and p3=0.286−p1+p5≥0p_3 = 0.286 - p_1 + p_5 \ge 0. So p1=0p_1 = 0 with p5=0.3p_5 = 0.3 is not a cohort, it gives p4=−0.283p_4 = -0.283, a negative share of students.

  2. Cohort A, no 1s and the largest possible block of 3s. Choose p1=0p_1 = 0 and p5=0.1585p_5 = 0.1585: then p4=0.317+0−0.317=0p_4 = 0.317 + 0 - 0.317 = 0, p3=0.603−0−0.1585=0.4445p_3 = 0.603 - 0 - 0.1585 = 0.4445, and p2=0.397p_2 = 0.397.

  3. Check A: mean =2(0.397)+3(0.4445)+5(0.1585)=0.794+1.3335+0.7925=2.92= 2(0.397) + 3(0.4445) + 5(0.1585) = 0.794 + 1.3335 + 0.7925 = 2.92, and the shares add to 0+0.397+0.4445+0+0.1585=10 + 0.397 + 0.4445 + 0 + 0.1585 = 1.

  4. Cohort B, nobody at a 3 or a 5. Choose p5=0p_5 = 0 and put the whole upper group at 4, so p4=0.603p_4 = 0.603 and p3=0p_3 = 0. Then p1=p4+2p5−0.317=0.603−0.317=0.286p_1 = p_4 + 2p_5 - 0.317 = 0.603 - 0.317 = 0.286 and p2=0.397−0.286=0.111p_2 = 0.397 - 0.286 = 0.111.

  5. Check B: mean =1(0.286)+2(0.111)+4(0.603)=0.286+0.222+2.412=2.92= 1(0.286) + 2(0.111) + 4(0.603) = 0.286 + 0.222 + 2.412 = 2.92, and the shares add to 0.286+0.111+0+0.603+0=10.286 + 0.111 + 0 + 0.603 + 0 = 1.

  6. Cohort C, an ordinary shape. Choose p1=0.16p_1 = 0.16 and p5=0.15p_5 = 0.15: then p4=0.317+0.16−0.30=0.177p_4 = 0.317 + 0.16 - 0.30 = 0.177, p3=0.603−0.177−0.15=0.276p_3 = 0.603 - 0.177 - 0.15 = 0.276, and p2=0.397−0.16=0.237p_2 = 0.397 - 0.16 = 0.237.

  7. Check C: mean =0.16+2(0.237)+3(0.276)+4(0.177)+5(0.15)=0.16+0.474+0.828+0.708+0.75=2.92= 0.16 + 2(0.237) + 3(0.276) + 4(0.177) + 5(0.15) = 0.16 + 0.474 + 0.828 + 0.708 + 0.75 = 2.92, and the shares add to 1.

  8. Cohort D, the most 5s the figures allow. Choose p1=0.397p_1 = 0.397, its ceiling, and then the largest p5p_5 that keeps p4=0.317+p1−2p5p_4 = 0.317 + p_1 - 2p_5 from going negative: 2p5≤0.317+0.397=0.7142p_5 \le 0.317 + 0.397 = 0.714, so p5=0.357p_5 = 0.357 and p4=0p_4 = 0. Then p3=0.603−0−0.357=0.246p_3 = 0.603 - 0 - 0.357 = 0.246 and p2=0.397−0.397=0p_2 = 0.397 - 0.397 = 0.

  9. Check D: mean =0.397+3(0.246)+5(0.357)=0.397+0.738+1.785=2.92= 0.397 + 3(0.246) + 5(0.357) = 0.397 + 0.738 + 1.785 = 2.92, and the shares add to 0.397+0+0.246+0+0.357=10.397 + 0 + 0.246 + 0 + 0.357 = 1.

  10. Compare the four. The share earning a 5 is 15.85%, 0%, 15.0% and 35.7%. The share earning exactly a 3 is 44.45%, 0%, 27.6% and 24.6%. The median is a 3 in A, C and D and a 4 in B. Every one of them reports a mean of 2.92 with 60.3% at 3 or higher.

All four cohorts publish identical headline figures and describe visibly different exams. A mean and a tail share constrain a distribution; they do not determine it. Any five-row table presented as the distribution behind those two numbers has added assumptions of its own.

Frequently asked questions

What is a good AP Statistics score?

It depends on the purpose. Measured against other test takers on the May 2025 results, 60.3% earned a 3 or higher and the mean was 2.92, so a 3 or better was the majority result and a 4 or 5 put a student in a group of at least 15.85% of test takers. Those two figures put the median at a 3 or a 4 without settling which. Measured against college credit, the only answer that counts is the threshold each college publishes for itself. Those May 2025 figures describe the old nine-unit course.

Is a 3 a good AP Statistics score?

On the May 2025 results, under the old nine-unit course, a 3 was at or above the middle: 39.7% of students scored below a 3 and 60.3% scored at it or above. That makes a 3 or better the majority result and puts the median at a 3 or a 4, so a 3 was a middle result rather than a weak one. Whether it is good enough for you is a different question, and it is answered by the credit policy of the colleges on your list, not by the distribution.

What percentage do you need for a 5 in AP Statistics?

There is no public answer. Composite cut points, the raw totals that map to each score from 1 to 5, are not published for the redesigned course, so any specific percentage you find has been invented. What is published is the weighting: 42 multiple-choice questions for 50% of the score, and four free-response questions worth 10 points each for the other 50%.

What is the average AP Statistics score?

The most recent figure recorded on this site is a mean of 2.92, from May 2025. That was the last administration under the previous nine-unit course. The first exam on the redesigned course is May 2027, so no mean exists for it yet, and one will not exist until after that exam is scored.

Will scores be lower on the redesigned 2027 exam?

Nobody knows, and this site will not guess. May 2027 is the first administration of the new course, sat by a new cohort under fully digital delivery in Bluebook. The sensible move is to plan against what is published, which is the exam structure and the unit weightings, rather than against a predicted curve.