How Grade Curves Actually Work

Every other page on this site answers from a document you are holding. This one cannot, and that is the honest starting point: “curved” names at least four different operations, your instructor almost certainly did not say which one they used, and the choice moves a score by more than a letter grade.

So the useful thing is not a formula. It is knowing what the four methods do, who each one helps, and how wide the range of honest answers is.

One raw score, four curves, three letters

Take a raw 68% in a class that averaged 66 with a spread of 12, where the top score was 88, the instructor mentioned adding 7 points, and the department targets a 77 average with a spread of 10. Every method below is defensible and every one is in use:

Three different letter grades, 7.5 percentage points apart, from one piece of work. Nothing about the exam changed. Only the arithmetic somebody chose in private.

Three of the four are the same operation

This is the finding that makes the list navigable. Write each method as a straight line — curved = intercept + slope × raw — and three of them collapse into one thing with a different part held still:

• Flat addition is that line with the slope pinned at 1. The whole class slides up together; the gaps between students are untouched.
• Scaling to the top score is the line with the intercept pinned at 0. The class stretches away from zero, so a high scorer gains more points than a low one.
• The mean-and-spread fit is the line with both free. It slides and stretches at once, to hit a target average and a target spread.

Mehvar (2025) arrives at the same reduction from the other direction, noting that when the ratio of adjusted to raw standard deviation is held at 1, the fit becomes a uniform shift — flat addition by another name. So the real question is never “which of four formulas”. It is: was the class slid, stretched, or reshaped?

Only the square-root curve is not a line, which is exactly why it behaves so differently in the table above — and why the folklore about it is wrong.

The square-root curve does least for the student who did worst

The method is stated everywhere as the one that helps struggling students most. Its arithmetic says otherwise. Curved score is 10 × √raw, so the gain is 10√raw − raw:

The gain peaks at a raw 25% and falls away on both sides. At a raw zero it is exactly nothing, because the square root of zero is zero. The student who did worst in the class gains the least of anyone — the opposite of the method's reputation, and not a close call.

What is true is that it is generous in the middle-low range, and that it never lowers a score and never pushes one past 100. It is also the only one of the four with no institutional publisher: no registrar or calendar defines it, and it circulates as classroom practice. That puts it in a different evidentiary class from the fit, which is documented in the literature.

A curve can take points away — and it is the mandatory kind that does

“Curved” is heard as “raised”. One of the four can subtract. The mean-and-spread fit pulls the class toward a target, and if your section's real spread is wider than the target, the top of the class comes down.

In the same class as above — a real mean of 66 and a spread of 12 — a raw 90 fitted to a 77 mean with a target spread of 4 comes back an 85. Five points gone, from the arithmetic doing exactly what it was asked to do.

That is not hypothetical. Michigan State University's College of Law publishes it as a requirement: a mandatory curve with a target mean of 3.00 for first-year classes and 3.33 for upper-level ones, tolerances of ±0.07 to ±0.17 depending on class size, and no more than 70% of any class in the B range. In a section that did unusually well, that policy is a subtraction, in writing, in advance.

Why nobody will tell you which one was used

Because at most institutions there is no rule to tell you. Michigan State's law school mandates a curve; the University of Alberta forbids mandating one at all — Grades in any course, examination or other academic assessment shall not be mandated on the basis of a curve or historic distribution of student grades. Between those two poles sit most universities, which publish nothing, because the choice is left with the instructor deliberately.

And many instructors curve by hand to no formula at all — moving a cut-off down a couple of points because question 7 was unfair. That is not any of the four methods and no calculator will reproduce it.

This is why the curve calculator shows all four side by side and never asks you to pick. Narrowing to one answer is the single thing the arithmetic gives no basis for. The range is the answer.

Where this stops

• A curve never reorders the class. Whatever method was used, if you scored above someone before, you still do. Curving changes the numbers, not the ranking.
• Class statistics you were not given cannot be worked out from your own score. Three of the four methods need the mean, the spread or the top score, and if your instructor did not publish those, those methods are genuinely unanswerable rather than answerable with an estimate.
• Moving the cut-offs is a different thing entirely.An instructor who announces “an 85 will be an A this term” has not curved anyone's score — they have changed the letter table, which is a separate lookup.
• Points and percentages are not the same units.“+5 points on a 60-point test” is 8.3 percentage points, and the conversion has to happen first.

Finally: whatever the curve did, it produced a score for one assessment. What that does to your course grade depends on the weight the assessment carries, which is the other calculation — and a generous curve on a 10% quiz is worth less than it feels.