Grade Curve Calculator
Paste a class's scores and see what a curve would actually do to them. Four methods, each a genuinely different operation, and for every one of them the count of A's, B's, C's, D's and F's before and after — which is the number you are really deciding on.
Curve the class
Paste or type up to 500 percentages, 0 through 100. Commas, spaces, tabs, and new lines all separate, so a gradebook column pastes straight in. Order does not matter — the result lists them back in the order you entered them.
The first three only ever raise a score. The square-root curve is the only nonlinear one: it rescues the bottom of the class and compresses the top. None of them fits your class to a bell curve — see the note below on why that is a different thing entirely.
Capping is the default and the usual gradebook behaviour. It has a side effect worth knowing: a cap flattens everyone it touches to the same 100, so a flat curve big enough to hit the ceiling stops being spread-preserving at the top.
Up from 72.07 across 15 papers — +4 points on every score. A grades go from 2 to 3, F grades from 3 to 2. No score reached the 100 ceiling.
- Curve applied
- +4 points to every score
- Papers curved
- 15
- Mean (before → after)
- 72.07 → 76.07
- Median (before → after)
- 72 → 76
- Lowest score (before → after)
- 45 → 49
- Highest score (before → after)
- 96 → 100
- Spread, σ (before → after)
- 14.23 → 14.23
- A grades (before → after)
- 2 → 3
- B grades (before → after)
- 3 → 3
- C grades (before → after)
- 4 → 4
- D grades (before → after)
- 3 → 3
- F grades (before → after)
- 3 → 2
- Scores held at 100
- 0
Four curves, and what each one moves
Every method here applies the same rule to every paper. Nobody gains at anyone else's expense and the ranking never changes — only where the class sits, and in one case how spread out it is.
flat: c = r + p • top-to-100: c = r + (100 − max) • root: c = √r × 10 • linear: c = r + (target − mean)- r
- The raw score, a percentage from 0 through 100
- c
- The curved score, floored at 0 and capped at 100 unless you allow more
- p
- The points you chose to add, for the flat curve
- max
- The highest raw score in the class
- mean
- The class average before the curve
Three of these move the class; one reshapes it
The flat, top-to-100 and linear methods all add a constant. That makes them shifts: the standard deviation of the class comes through untouched, every gap between two students stays exactly the width it was, and the only thing that changes is where the whole distribution sits. In the default class above, the spread is 14.23 points before the curve and 14.23 points after it.
The square-root curve is the exception, and the only nonlinear method on this page. It replaces each score with √score × 10, which maps 0–100 onto 0–100 — a 100 stays a 100, so the top of the scale is never broken — but it does so unevenly. The lift is largest at the bottom and shrinks all the way up. That is the feature and the cost in one property, and the section below spells out the arithmetic.
What this calculator will not do
It will not fit your class to a normal distribution. It will not stretch the spread — a z-score rescale changes the gaps between students, not just where the class sits, and it needs a target spread nobody has an intuition for. And it will not drop, weight or replace individual scores; the input is the list of marks as they stand.
“Curving” means two different things, and only one of them is on this page
In ordinary use, curving a test means any upward adjustment. Everyone got five points; the top paper was raised to 100 and the rest came with it. Nobody hearing it in a staffroom thinks about distributions at all. That colloquial meaning is what this tool implements.
The term did not start there. Grading on a curve meant fitting the class to a normal distribution — the curve is the bell — so that a fixed share of students received each letter no matter how the class performed. The version usually credited to the psychologist Max Meyer, published in Science in 1908, assigned the top 3% an A and the bottom 3% an F, with half the class landing on a C.
Its defining property is the one it is now widely criticised for: the proportions are decided before the test is marked. Picture a class where every score falls between 88 and 96 — everyone learned the material. Fit that class to a quota with a 10% failure band and the 88 fails, for work that is a solid B against any absolute standard. Run the same quota on a class that collectively struggled and it still produces its allotted A's. The grade stops describing what a student knows and starts describing where they placed in one particular room.
That is also why it changes the incentives inside a classroom. If A's are rationed, helping the person next to you costs you something. Norm-referenced scoring has a legitimate home — it is how standardised tests report percentiles, where ranking is the explicit purpose — but a course grade is normally meant to be criterion-referenced, a claim about whether the student met the standard. This calculator stays on the criterion side of that line: every method is a transformation applied identically to every paper, and no student's letter depends on anyone else's.
One class of fifteen, run through all four curves
The scores loaded above are a believable set: 45, 52, 58, 61, 63, 67, 70, 72, 74, 78, 81, 85, 88, 91, 96. The average is 72.07, the median is 72, and the spread is 14.23 points. On the straight ten-point scale that is 2 A's, 3 B's, 4 C's, 3 D's and 3 F's.
Highest score to 100
The best paper is a 96, so everyone gains 100 − 96 = 4 points. The average moves to 76.07 and the spread stays at 14.23 — a pure shift. In letters: the 88 crosses into an A and the 58 crosses out of an F, so the class becomes 3 A's, 3 B's, 4 C's, 3 D's and 2 F's. One student rescued, one promoted, everyone else in the same band they started in. If four points is not the adjustment you meant to make, this is where you find that out.
Flat points
Add 5 to everyone and the average lands on 77.00 — not 77.07, because the 96 would become a 101 and is held at 100. That single cap is worth watching. Add 20 instead and five separate papers — the 81, 85, 88, 91 and 96 — all land on exactly 100, which means five students who were 15 points apart now share a grade. A flat curve preserves the spread right up until the ceiling, and then it stops.
Square root
√45 × 10 = 67.08. √72 × 10 = 84.85. √96 × 10 = 97.98. The average jumps to 84.46 and every F disappears: the distribution becomes 5 A's, 5 B's, 4 C's, 1 D and no F's.
Now look at what it cost. The spread falls from 14.23 to 8.54. The gap between the top paper and the median paper — 96 against 72, twenty-four points of difference — closes to 97.98 against 84.85, or about thirteen points. The student who scored 45 gains 22 points; the student who scored 91 gains 4. That is not a side effect, it is the method: √score × 10 is steep at the bottom and flat at the top, so it buys its rescue of the failing students with the distance between your strong ones and your average ones. Whether that is the right trade depends on why the class scored badly, and it is a judgement a teacher should make on purpose rather than discover in the gradebook.
Linear shift to a target average
Ask for an average of 80 and every score gains 80 − 72.07 = 7.93 points. The spread is again 14.23, unchanged, because a shift cannot change it. This is the method to use when a department has a number to hit, and the only one that will move grades down if the target sits below where the class already is.
What this calculator assumes
- Scores are percentages from 0 through 100. If your test is out of 40 points, convert to a percentage first. Extra credit above full marks is out of scope: it would make the top-to-100 method subtract points from the class and make the square-root curve run downhill, and neither of those is a curve.
- Letters are read off the straight ten-point scale — A from 90, B from 80, C from 70, D from 60. It is the commonest convention in the US, not a standard, and it is fixed here so the before and after counts are measured against the same ruler.
- Curved scores are capped at 100 unless you switch that off, and the result reports how many were held there. Nothing is capped silently.
- No score ever goes below 0. A downward linear target large enough to push scores negative will therefore land the class above the average you asked for, and the result panel shows the average that actually came out.
- The spread is the population standard deviation, dividing by n. A class is the whole group being described, not a sample drawn from a larger one.
- Nothing is fitted to a distribution and nothing is ranked. Every method applies the same rule to every paper, so no student's curved grade depends on another student's raw one, except through the single class statistic — the maximum, or the mean — that you chose to anchor the curve on.
- Figures are rounded to four decimal places internally and displayed to two. The letter is read off the rounded score, so a number shown as 90 is never labelled a B.
Grade curve FAQ
What does “grading on a curve” actually mean?
Two different things, and the gap between them matters. In everyday staffroom use it means any upward adjustment: everyone got five points, or the top paper was raised to 100 and the rest came with it. That is what this calculator does. The original meaning was narrower and stronger — fitting the class to a normal distribution, the bell curve the phrase names, so that a fixed share of students received each letter. The scheme usually credited to the psychologist Max Meyer, published in Science in 1908, gave the top 3% an A, the bottom 3% an F and half the class a C. Its defining feature is that those proportions are set before anyone marks the test.
Why won't this tool fit my class to a bell curve?
Because the fixed proportions are the whole problem. Imagine a class where every score falls between 88 and 96 — everybody learned the material. Fit that class to a quota with a 10% failure band and the 88 gets an F, for work that was a solid B on any absolute standard. The reverse holds too: a class that collectively struggled still produces its allotted A's. A quota measures students against each other rather than against what they were supposed to learn, which is why it has drifted out of favour in school grading even though it remains standard practice in norm-referenced testing, where ranking is the explicit point. It also makes classmates competitors for a rationed grade. Every method here instead applies the same transformation to every paper, so nobody's letter depends on how anyone else did and nobody's rank changes.
Which curve should I use?
Ask what went wrong. If a specific question was unfair or ungradeable, the honest fix is flat points equal to what that question was worth — it repays exactly the damage. If the test was simply harder than intended and the best paper in the room represents the ceiling of what was achievable, raising the highest score to 100 says so. If you have a departmental target average to hit, the linear shift moves the class onto it without touching the spread. Reach for the square root only when the bottom of the distribution is the problem you are solving, because it is the only method here that changes the shape of the class rather than its position.
Why does the square-root curve help a 45 so much more than a 91?
Because √score × 10 is a curve in the geometric sense — it is steep at the bottom and flat at the top. A 45 becomes a 67, a gain of 22 points. A 91 becomes a 95, a gain of 4. A 100 gains nothing at all, which is what makes it usable: it maps the 0–100 range exactly onto itself. The consequence is that it compresses the top. In the default class here the gap between the 96 and the 72 falls from 24 raw points to 13 curved ones. If you believe a strong paper and an average one should stay visibly apart, this curve works against you; if you believe the test was too punishing at the low end, it is doing precisely what you want. Decide which, rather than discovering it afterwards.
Can a curve lower someone's grade?
Only one setting here can, and only if you ask for it. Flat points, highest-score-to-100 and the square root can never reduce a score. Shifting the class onto a target average will lower every score if the target sits below the current average — that is arithmetic, not a bug, and it is the reason the option exists for departments that have to hit a number in both directions. Scores never go below 0, so a large downward shift can leave the resulting average above the target you asked for. The result panel prints the average that actually came out, not the one you requested.
What happens to a curved score above 100?
It is capped at 100 by default, and the result says how many scores were held there. That matches most gradebooks, but it has a side effect worth naming: a cap flattens everyone it touches to the same number, so a curve big enough to hit the ceiling stops preserving the spread at the top and starts erasing the distinction between your best students. Add 20 points to the default class here and five different papers all land on exactly 100. Switch the setting to allow scores past 100 and you can see the uncapped figures instead.
Which letter scale does the distribution use?
The straight ten-point scale — A from 90, B from 80, C from 70, D from 60, F below that. It is the most widespread convention in United States schools, not a national standard, because there is no national standard. A school that starts its A at 93, or uses plus and minus bands, will read one or two of these students differently. The scale is fixed here on purpose: the question this tool answers is whether the curve moved people between letters, and that comparison only means something if both sides are read against the same ruler.
Sources and review notes
- AERA, APA and NCME — Standards for Educational and Psychological Testing, the standing reference for the distinction between norm-referenced and criterion-referenced score interpretation that this page turns on
- NCES Digest of Education Statistics — the federal reference for how grades are recorded across US schools, and the evidence that there is no national grading scale to appeal to
- Max Meyer, “The Grading of Students”, Science, 1908 — the origin of the fixed-proportion scheme (3% A, 22% B, 50% C, 22% D, 3% F) that the phrase “grading on a curve” originally named
The arithmetic on this page is definitional and cannot go stale. What is a judgement rather than a fact is the ten-point letter scale, which varies school by school, and the choice of curve itself, which is pedagogy and not mathematics. The engine behind this page is a pure function under unit test, including a case that fails if an additive curve ever stops preserving the spread.