Free ESAT guide, no signup, sources named and linked

What is a good ESAT score?

For the October 2024 cohort, 4.5 was the median and 7.0 the 90th percentile in every module. Across both sittings, module medians were 4.2 to 4.5. There is no pass mark, no total score, and no published Cambridge or Imperial offer-holder average.

Download the PDFChecked August 2026About 8 minutes to read

First, bin the idea of an ESAT total

You receive a separate scaled score from 1.0 to 9.0, reported to one decimal place, for every module you sit. The modules are equated and scaled separately. A 5.8 in Physics and a 5.8 in Biology are not interchangeable measurements, and adding three module scores creates a number the test itself does not recognise.

The score is not simply “marks out of 27, stretched to nine”. UAT-UK estimates ability while allowing for small differences between test forms, then transforms that estimate onto the 1.0 to 9.0 scale. Its technical report does not print a raw-mark conversion table. An exact calculator that says “17/27 always equals 6.2” is claiming data that UAT-UK has not released.

The published distribution

The first technical report covers 11,919 candidates applying for 2025 entry. UAT-UK fixed the October ability median to 4.5 and the October 90th percentile to 7.0, module by module. January used the same scaling constants, so its lower results pulled some combined figures down.

Combined October 2024 and January 2025, Table 5
Module	N	25th	Median	75th	90th
Mathematics 1	11,919	3.3	4.2	5.4	6.7
Mathematics 2	10,040	3.4	4.5	5.5	6.8
Physics	9,237	3.4	4.4	5.6	6.8
Chemistry	2,821	3.4	4.5	5.7	6.8
Biology	1,429	3.6	4.5	5.9	7.0

Around 4.5 is the middle, with Mathematics 1 slightly lower at 4.2 across both sittings.
Roughly 5.4 to 5.9 reaches the top quarter, depending on the module.
Roughly 6.7 to 7.0 reaches the top tenth. In October alone, 7.0 was the 90th percentile in all five modules.

So what should you call good?

Above the median: evidence that you outperformed more than half of an already self-selecting admissions-test cohort.
Top quarter: a clearly strong module result, around the mid-to-high fives in the published cycle.
Top tenth: an exceptional cohort position, around 7.0 in the published cycle.

That still does not tell you what a particular course wants. UAT-UK states there are no grade boundaries or pass mark. Universities read the result alongside the rest of the application. A score can strengthen an application without guaranteeing an interview; a weaker score can hurt without acting as an automatic rejection switch.

Scaled scores also have measurement error. The report gives module standard errors of roughly 0.67 to 0.97 scaled points, depending on module and sitting. Treat 5.9 and 6.1 as nearby evidence, not as two different species of applicant.

What do Cambridge and Imperial offer-holders score?

The current primary sources do not publish an offer-holder distribution or average for either university. They also do not publish course cut-offs. There is no defensible sentence of the form “Cambridge Engineering offer-holders typically get X” yet.

Cambridge applicants must use the October sitting. Its 2024 cohort had a 4.5 median and 7.0 90th percentile in every module, but included several universities and application outcomes. It is not an offer-holder table.
Imperial says it considers ESAT and TMUA scores alongside the UCAS application when deciding interviews and offers. It does not publish a pass mark or offer-holder average.

Reason modestly: above-median is positive cohort evidence, top-quarter is stronger, and top-tenth is rarer. Calling 6.5 “the Cambridge average” would turn an inference into a fake statistic.

One mark is not one score step

Worked example A parameter trap from Mathematics 2

The equation $x^2+(k-3)x+k=0$ has no real roots. Which values of $k$ work?

No real roots means the discriminant is strictly negative:

(k-3)^2-4k<0.

Expand and factorise:

k^2-10k+9<0 \Longleftrightarrow (k-1)(k-9)<0.

An upward-opening quadratic is below zero between its roots:

\boxed{1<k<9}.

Correct: the endpoints are excluded because a zero discriminant gives one repeated real root.

Trap: $1\leq k\leq9$ silently changes “no real roots” into “at most one root”.

That question earns one raw mark. Its scaled value is not fixed in advance. It depends on the complete response pattern, form difficulty and post-test equating. Train for marks; interpret the final scaled score with percentiles.

A target that survives contact with reality

Below the cohort median, chase reliable first-pass marks and remove blanks.
Around the median, identify the three question types consuming two minutes each.
Near the top quarter, review errors by cause: missing knowledge, slow execution, misread wording, or an untested distractor.

A 7.0 target does not mean “get every hard question”. A top-tenth score still belongs to a paper designed to challenge extremely able candidates. Aim for a clean first pass, disciplined banking and zero avoidable blanks.

Sources

UAT-UK ESAT Technical Report 2024-25: scaling, Table 5 distributions, reliability and standard error.

UAT-UK test results: separate module scores and no grade boundaries or pass mark.

Imperial score guidance: scores are considered alongside the application, without a published pass mark or offer-holder average.

Keep going, still free

All ESAT guides Timing gym Free diagnostic