The exams come back, you average them, and the mean is 58%. The material was taught, the questions were fair — the test was just hard. Now what? “Curving” covers at least five different operations, and they are not interchangeable. Each moves points differently, and picking one is a fairness decision whether you name it or not.

Method 1 — Flat add: everyone gets the same points

Add a fixed number to every score: new = old + 10. If the test was uniformly 10 points too hard, this is the honest fix — it preserves every gap between students exactly.

Watch for: the ceiling. Anyone within 10 points of 100 needs a cap (or “extra credit” accounting), and the method does nothing about the fact that a 10-point gift means proportionally more to a failing score than a passing one.

Method 2 — Anchor to the top score

Find the highest score, then either add the gap to everyone (new = old + (100 − max)) or scale everyone by it (new = old × 100 ÷ max). If the best student hit 88, either everyone gets +12, or every score multiplies by 1.136.

Watch for: the outlier problem. One exceptional student sets the whole class’s curve. If the top score is 96 and second place is 81, anchoring to 96 barely helps anyone; anchor to the 90th percentile instead and the curve reflects the class, not the genius.

Method 3 — Square root: help the bottom most

new = 10 × √(old), with scores as percentages. The math has a beautiful property: it lifts low scores enormously and barely touches high ones.

Raw	Flat +10	Top-anchor (93)	√-curve	Points-back ½
45	55	52	67.1	72.5
58	68	65	76.2	79.0
70	80	77	83.7	85.0
82	92	89	90.6	91.0
93	100	100	96.4	96.5

A 45 becomes a 67 — failing to passing — while the 93 stays a high A. For a test that was simply pitched too high, this is usually the fairest single move.

Method 4 — Points back: give everyone half the distance to 100

new = old + (100 − old) ÷ 2. A 45 gains 27.5 points, a 93 gains 3.5. It compresses the whole distribution toward the top — gentler than square root for mid scores, more generous than flat add for low ones. Useful when you want recovery without a pile of new A+s.

Method 5 — The actual bell curve (proceed with care)

Strictly, “grading on a curve” means assigning letters by rank: top ~10% get A’s, next ~20% B’s, and so on down a forced distribution. It answers a different question — “who learned most relative to the class?” — and it’s how some law schools and curved STEM courses work by design.

Watch for: it’s a zero-sum game. Students compete against each other rather than against the material; a strong class pushes good scores into C’s, a weak class inflates mediocre ones. Unless your institution mandates it, a rank-based curve on a too-hard test punishes students twice.

Before you curve at all

Two things beat every formula. First, audit the items: if a question was ambiguous or mis-keyed, rescore it for everyone — that fix is correction, not curving, and it should happen before any curve is applied. Second, consider whether the test measured what you taught; if the answer is no, a retest teaches more than a rescale.

And whatever method you pick: publish it. Students who can see that 10√(old) moved their 58 to a 76 trust the curve; students who watch numbers change for no stated reason learn that grades are weather — arbitrary, and out of their control.

When the curve is decided, the grading chart handles the rest — set the question count, and the scale your syllabus uses does the lettering.

Frequently asked questions

Is curving a test the same as grading on a bell curve?

No — and the confusion causes real arguments. 'Curving' usually means adjusting every score upward by some rule. 'Grading on a curve' in the strict sense means assigning letters by rank so a fixed share of students gets each grade — the bell curve. The first lifts everyone; the second pits students against each other and can lower a good score's letter.

What's the fairest curve for a test that was too hard?

The square-root curve is the usual answer: new score = 10 × √(old score). It lifts low scores the most (a 45 becomes a 67) while barely moving high ones (a 93 becomes a 96), so it fixes a too-hard test without compressing the top of the class into indistinguishable A+s.

Should the top score set the curve?

It's popular because it's self-calibrating — if the best student got 88, the test was at least 12 points too hard. The risk: one outlier (a genius or a lucky guesser) sets the whole class's curve. If the second-best score is far below the top, anchor to the 90th percentile instead of the single maximum.

Can a curve ever lower someone's grade?

With flat-add, top-anchor and square-root methods, never — they only move scores up or leave them flat. A true bell curve absolutely can: under forced distribution, a raw 85 can land a C if the class was strong. If any method you're considering moves a score downward, you're not curving — you're re-normalizing.

Is it better to curve or to throw out bad questions?

Fix the instrument first. If an item was ambiguous or mis-keyed, rescore it for everyone — that's correction, not curving. Curves are for the residue: the test was fair but too hard, so the distribution sits lower than the mastery it measured. Curving to compensate for three broken items hides the problem and still mis-grades the students who lost points on them.