How Are Grade Thresholds Set?

A student-friendly guide to understanding Cambridge exam grading

What you'll learn

  • What grade thresholds actually are
  • Why they change every exam session
  • How Cambridge decides on the numbers
  • What this means for your exam preparation

What is a Grade Threshold?

After every exam session, Cambridge publishes a table showing the minimum marks needed for each grade.

Example: Physics Paper 4 (May 2024)

A*
67/80
A
58/80
B
49/80
C
40/80
D
32/80

If you scored 58 or above, you got an A. If you scored 67 or above, you got an A*.

Why Do Thresholds Change?

Here's something that surprises many students:

There is no fixed percentage for any grade.

An A* is NOT always 90%. It could be 75% in one session and 88% in another.

Why? Because some exam papers are harder than others.

Imagine two students with the same ability:

Session 1: Hard Paper

Student scores 65/80

Threshold for A: 58/80

Result: Grade A ✓

Session 2: Easy Paper

Student scores 72/80

Threshold for A: 70/80

Result: Grade A ✓

Both students got an A, even though their marks were different. The threshold adjusted for the paper difficulty.

This is called "comparable outcomes" — making sure a grade means the same thing regardless of which session you sat.

How Does Cambridge Decide the Thresholds?

Cambridge uses a three-step process:

1

Analyze the marks

They look at how everyone performed — the average mark, the spread of marks, and how it compares to previous years.

2

Review actual exam scripts

Senior examiners read scripts at the proposed grade boundaries. They ask: "Does this work deserve an A? Does this deserve a B?"

3

Committee decision

A committee reviews all the evidence and sets the final thresholds.

Important note

The exact formulas Cambridge uses are not public. Anyone claiming to know the "secret formula" is guessing.

A Simple Example of How It Might Work

Disclaimer: This is a simplified example for learning purposes. We don't know if Cambridge uses this exact method.

One way to adjust for paper difficulty:

Step 1: Calculate how hard the paper was
Historical average: 55/80
This year's average: 45/80 (harder!)
Difficulty factor: 45 ÷ 55 = 0.82

Step 2: Adjust the threshold
Normal A threshold: 65/80
Adjusted: 65 × 0.82 = 53/80

Because the paper was harder, the threshold dropped from 65 to 53.

Don't use this to predict thresholds!

The real process involves expert judgment, not just math. This is just one piece of the puzzle.

Do Other Exam Boards Do the Same?

Yes! Most major exam boards use similar approaches:

Edexcel, AQA, OCR (UK boards)

All follow Ofqual's "comparable outcomes" rules

Cambridge (CAIE)

Uses comparable outcomes for international students

~

IB Diploma

Slightly different — more focused on meeting specific criteria

What Does This Mean for You?

✓ DO: Aim higher than you think you need

If past A* thresholds ranged from 70-85%, aim for 90%+. Give yourself a safety margin.

✓ DO: Focus on understanding, not thresholds

Students who truly understand the material don't need to worry about where thresholds land.

✗ DON'T: Assume past thresholds will repeat

Every paper is different. Last year's thresholds don't predict this year's.

✗ DON'T: Try to "game" the system

There's no shortcut. Solid preparation beats threshold-watching every time.

Key Takeaways

  1. Thresholds are not fixed — they adjust based on paper difficulty
  2. Cambridge uses data + expert judgment — the exact formula is not public
  3. All major exam boards do this — it's standard practice
  4. Your best strategy: Master the content and aim high

Where to Find Thresholds

Cambridge publishes thresholds a few weeks after results:

Deep Dive: Statistical Methods for Adjusting Thresholds

For those who want to understand the math behind grade adjustments

Important: These are statistical methods that COULD be used. We don't know which method(s) Cambridge actually uses. This section is for educational purposes only.

1. Difficulty Factor Method

The simplest approach: adjust thresholds proportionally to paper difficulty.

Difficulty_Factor = This_Year_Mean ÷ Historical_Mean
Adjusted_Threshold = Historical_Threshold × Difficulty_Factor

Worked Example:

• Historical mean: 55/80

• This year's mean: 45/80 (harder paper)

• Difficulty factor: 45 ÷ 55 = 0.82

• Historical A threshold: 65/80

• Adjusted A threshold: 65 × 0.82 = 53/80

Pros: Simple to calculate. Cons: Doesn't account for cohort ability differences.

2. Percentile Matching

Keep the same percentage of students achieving each grade every year.

If historically X% get Grade G:
→ Find mark where top X% scored above
→ That mark becomes the threshold for G

Worked Example:

• Total candidates: 10,000

• Historical A* rate: 8%

• Top 8% = 800 students

• Sort all marks highest to lowest

• Find 800th highest mark: 67/80

• A* threshold = 67

Pros: Maintains consistent grade distribution. Cons: Ignores if this year's cohort is actually stronger/weaker.

3. Standard Deviation (Z-Score) Method

Set thresholds at fixed standard deviations from the mean.

Threshold = Mean + (Z × Standard_Deviation)

Where Z is a fixed value for each grade:
A* → Z = 1.5 | A → Z = 1.0 | B → Z = 0.5 | C → Z = 0

Worked Example:

Mean = 50/80, Standard Deviation = 12

A* (Z=1.5): 50 + (1.5 × 12) = 68
A (Z=1.0): 50 + (1.0 × 12) = 62
B (Z=0.5): 50 + (0.5 × 12) = 56
C (Z=0.0): 50 + (0.0 × 12) = 50

Pros: Automatically adjusts to paper difficulty. Cons: Assumes normal distribution.

4. Prior Attainment Regression

Predict expected grades based on students' previous exam performance.

1. Collect prior data (GCSE, mock results, etc.)
2. Build model: Expected_Mark = a + b × Prior_Mark
3. Predict grade distribution for this cohort
4. Set thresholds to match predicted distribution

Worked Example:

• This cohort's average prior attainment: Higher than last year

• Predicted A* rate: 10% (vs 8% historically)

• Actual A* rate at threshold 67: 10%

• Threshold stays at 67 (cohort is genuinely stronger)

This is likely closest to what Cambridge actually uses. It accounts for both paper difficulty AND cohort ability.

5. Equipercentile Equating

Match score distributions between different years/papers.

Find scores that represent the same percentile:
Score_2024 is equivalent to Score_2023
if both are at the same percentile rank

Worked Example:

• 2023: Score of 70 = 85th percentile

• 2024: Score of 65 = 85th percentile

• Therefore: 65 in 2024 ≈ 70 in 2023

• If A threshold was 70 in 2023 → it's 65 in 2024

Pros: Directly compares papers. Cons: Requires same population assumptions.

6. Anchor Items Method

Include some identical questions across years to directly measure difficulty change.

Difficulty_Change = Anchor_Mean_This_Year ÷ Anchor_Mean_Last_Year
Adjust all thresholds by this factor

Worked Example:

• Questions 15-20 are identical to last year

• Last year's average on Q15-20: 18/24

• This year's average on Q15-20: 15/24

• Difficulty factor: 15 ÷ 18 = 0.83

• This year's paper is harder → lower thresholds by 17%

Pros: Direct measurement of difficulty. Cons: Requires pre-tested questions.

7. Item Response Theory (IRT)

Sophisticated psychometric model used in major standardized tests (SAT, GRE).

P(correct) = 1 / (1 + e^(-a(θ - b)))

Where:
θ = student ability (estimated)
b = question difficulty parameter
a = question discrimination parameter

Key Concepts:

Difficulty (b): How hard is the question? Higher b = harder question.

Discrimination (a): How well does the question separate strong from weak students?

Ability (θ): Student's underlying ability, estimated from all their responses.

Grades are based on ability estimates, not raw scores. This means getting an easy question wrong hurts more than getting a hard question wrong.

Pros: Most accurate and fair. Cons: Requires extensive question pre-testing and complex computation.

Summary: Comparing All Methods

Method Complexity Data Needed Best For
Difficulty FactorMean onlyQuick estimates
PercentileAll marksFixed grade rates
Z-Score⭐⭐Mean + SDNormal distributions
Prior Attainment⭐⭐⭐Student historyFairest outcomes
Equipercentile⭐⭐All marksYear comparisons
Anchor Items⭐⭐Common QsDirect difficulty measure
IRT⭐⭐⭐⭐All responsesLarge-scale testing

What does Cambridge likely use?

Most likely a combination of Prior Attainment + Expert Script Review. The statistical prediction provides a starting point, and senior examiners refine it by reviewing actual student work at the boundaries.

Further Reading


This guide is for educational purposes. For official information, refer to Cambridge Assessment International Education.