EPSO numerical reasoning test: how it works, and how to beat the clock
The EPSO numerical reasoning test does not measure how good you are at maths. It measures whether you can find the two numbers that matter in a crowded table, choose the right operation, and stop calculating at the moment the answer options stop disagreeing. That is a method, and a method can be trained.
Short answer
The EPSO numerical reasoning test presents a table or chart of data and asks a multiple-choice question about it, with five options of which one is correct. In recent competitions it has typically been 10 questions in about 20 minutes — roughly two minutes each — and a basic on-screen calculator is provided.
The arithmetic never goes beyond percentages, ratios and division. The difficulty is finding the two numbers that matter in a crowded table fast enough, and putting the right value on the bottom of the fraction.
The most reliable way to prepare is short daily practice over six to eight weeks across the five recurring question types, using corrections that show the reasoning step by step rather than just naming the right option. Speed here comes from recognising the question type in seconds, and recognition is built by spacing practice out rather than cramming it.
| Test | Numerical reasoning, part of the EPSO computer-based reasoning battery |
|---|---|
| Format | Multiple choice — one data table or chart, one question, five options |
| Typical length | About 10 questions in about 20 minutes |
| Time per question | Roughly 120 seconds |
| Calculator | Basic on-screen calculator provided; you cannot bring your own |
| Note-taking | Depends on delivery: pen and paper at a test centre, an on-screen scratchpad only when remotely proctored |
| Delivery | At an accredited test centre or remotely proctored — your invitation says which |
| Wrong answers | No penalty — a blank scores the same as a wrong answer, so never leave one |
| Maths level required | Percentages, ratios, division — nothing above secondary school |
| Marking | Often combined with abstract reasoning against a single threshold |
| Recommended preparation | 6 to 8 weeks of short daily sessions |
| Binding source | The Notice of Competition for your own profile |
What the EPSO numerical reasoning test actually is
Numerical reasoning is one of the three computer-based reasoning tests used by the European Personnel Selection Office to filter candidates for EU institution competitions, alongside verbal reasoning and abstract reasoning. You sit it on a screen at an accredited test centre, and it is multiple-choice throughout: one table or chart of data, one question, five answer options, exactly one of which is right.
In recent competitions the numerical block has typically been 10 questions in about 20 minutes — around two minutes each. Treat those numbers as the shape of the thing rather than as gospel: the number of questions, the time allowed and the pass mark are fixed by each Notice of Competition, they differ between AD, AST and AST-SC profiles, and they have changed from one competition year to the next.
Every figure on this page describes what candidates have typically met in recent years. Before you build a pacing plan, open the Notice of Competition for your own profile on the EU careers site and read the section on the reasoning tests. If it disagrees with anything here, it wins.
What is provided, and what is not
- A basic on-screen calculator. Four operations, nothing more. You cannot bring your own. Reaching for it with the mouse costs three or four seconds each time, which is why estimation beats typing on most questions.
- Somewhere to write things down — but check which. At an accredited test centre you are given a pen and a few sheets of paper. Under remote proctoring you are not: physical paper and pen are forbidden, and you get an on-screen scratchpad instead. Your invitation states how your session is delivered.
- No penalty for a wrong answer in the standard multiple-choice format — so never leave a question blank. With five options, a blind guess is worth 0.2 marks and a blank is worth zero.
This is the detail most preparation skips, and it is worth more than another fifty questions. Typing an intermediate result into an on-screen scratchpad is slower and clumsier than writing it, and a candidate who has spent six weeks scribbling on paper discovers that for the first time on the day.
If your session is remotely proctored, do at least the final two weeks of timed practice with no paper on the desk — screen only. If it is at a test centre, keep the pen. Either way, decide which one you are training for before week eight rather than after.
The five question types you will actually meet
Across every EPSO numerical reasoning paper the questions collapse into a small number of families. Learning to name the family within the first ten seconds is most of the battle, because the family tells you which two numbers to look for.
Direct lookup
The number is in the table; the work is reading the right row and column, under time pressure, with similar labels designed to be confused.
Missing data
The number you are asked for is not in the table. You are given a part and a percentage, and must recover the whole. Cross-multiplication.
Percentage change
Growth, decline, share of total. The trap is always which value goes on the bottom of the fraction — and whether the question says percent or percentage points.
Compound change
Two successive rises or falls. They never add up. A 25% rise followed by a 20% fall lands you exactly where you started.
Ratios and rates
Per capita, per employee, per square kilometre; which country ranks highest. Usually answerable by comparison alone, without a single full division.
Free practice question 1 — the number that is not in the table
| Country | R&D as % of GDP | R&D spend (€m) |
|---|---|---|
| Belgium | 1.97 | 3,900 |
| Finland | 3.34 | 3,725 |
| Sweden | 3.78 | 7,630 |
| Portugal | 0.73 | 890 |
What was Finland's GDP in 2024, to the nearest million euro?
- A €89,300m
- B €111,527m
- C €124,400m
- D €197,600m
- E Cannot be determined
Show the worked solution
Answer: B — €111,527m
GDP is nowhere in the table, and option E is there to tempt you into thinking that settles it. It does not: you have a part (€3,725m of spending) and the percentage that part represents (3.34% of GDP). One unknown, one equation.
3,725 = 3.34% × GDP → GDP = 3,725 ÷ 0.0334 = 111,527The ten-second check. Before dividing, sanity-check the order of magnitude: 3.34% is roughly one thirtieth, so GDP must be roughly thirty times 3,725, i.e. around 110,000. Only option B is in that region. On a real paper you can mark B and move on without ever finishing the division.
Why the distractors exist. A is what you get by dividing by the wrong country's percentage. C and D are plausible-looking magnitudes for a country of Finland's size — they punish guessing by feel rather than by estimate.
Free practice question 2 — percent, or percentage points?
| Country | 2019 | 2024 |
|---|---|---|
| Spain | 14.1 | 11.3 |
| Greece | 17.3 | 10.1 |
| Italy | 9.9 | 7.6 |
| Ireland | 5.0 | 4.3 |
By what percentage did the Greek unemployment rate fall between 2019 and 2024?
- A 7.2%
- B 41.6%
- C 58.4%
- D 71.3%
- E 17.3%
Show the worked solution
Answer: B — 41.6%
A percentage change is always measured against the starting value. Here that is the 2019 figure, 17.3.
(17.3 − 10.1) ÷ 17.3 = 7.2 ÷ 17.3 = 0.416 → 41.6%Every distractor is a real mistake. This question is a museum of the four ways candidates lose this mark:
- A — 7.2% is the fall in percentage points, reported as if it were a percentage. The question said "by what percentage", and those are different quantities.
- D — 71.3% divides by the wrong number: 7.2 ÷ 10.1. The base is the value you started from, not the one you ended on.
- C — 58.4% is 10.1 ÷ 17.3, the share that remains — the complement of the answer, not the answer.
- E — 17.3% is simply a figure lifted out of the table. There is one of these in almost every question.
A rate moving from 17.3% to 10.1% has fallen by 7.2 percentage points and by 41.6 percent. Both statements are true and they are not the same statement. Read which one the question asked for — that word choice is deliberate.
Free practice question 3 — two changes that do not add up
A programme budget line stood at €48m in 2022. It increased by 25% in 2023, then decreased by 20% in 2024. What was it in 2024?
- A €45.6m
- B €48.0m
- C €50.4m
- D €52.8m
- E €60.0m
Show the worked solution
Answer: B — €48.0m, exactly where it started
Percentages do not add, they multiply. Convert each change into a multiplier and chain them:
48 × 1.25 = 60 60 × 0.80 = 48The 20% fall is taken on €60m, not on the original €48m, and 20% of 60 is exactly the 12 that the rise had added. The two cancel.
C — €50.4m is the trap, and it catches a lot of people: it is what you get by adding the percentages (+25 − 20 = +5) and applying 5% to 48. That reasoning is wrong every single time the two changes are of different sizes or opposite signs.
A rise of x% followed by a fall of x% never returns to the start — it always ends lower. +25% then −25% leaves you at 93.75% of where you began. The only reason this question lands back on 48 is that the percentages differ.
Free practice question 4 — the division you should not do
| Country | Spend (€m) | Population (m) |
|---|---|---|
| Ireland | 1,240 | 5.1 |
| Denmark | 1,690 | 5.9 |
| Finland | 1,480 | 5.6 |
| Norway | 1,850 | 5.5 |
Which country spent the most per inhabitant?
- A Ireland
- B Denmark
- C Finland
- D Norway
- E Cannot be determined
Show the worked solution
Answer: D — Norway
The slow route is four divisions: 243, 286, 264 and 336 euro per inhabitant. Correct, and it costs you ninety seconds you do not have.
The fast route is a comparison. Norway has the largest numerator of the four and the second-smallest denominator. A larger top over a smaller bottom is a larger fraction, so Norway beats Denmark, Finland and Ireland without any arithmetic at all. Answer, five seconds, next question.
Whenever a question asks which is highest rather than what is the value, look for a row that dominates on both axes before you divide anything. On a properly set paper there is usually one — and when there is not, you only need to divide the two or three candidates that remain, not all five.
Free practice question 5 — index points are not percent
| Country | 2015 | 2020 | 2024 |
|---|---|---|---|
| Germany | 100.0 | 108.4 | 124.6 |
| France | 100.0 | 106.2 | 119.8 |
| Poland | 100.0 | 112.7 | 148.3 |
By what percentage did consumer prices rise in Poland between 2020 and 2024?
- A 24.0%
- B 31.6%
- C 35.6%
- D 48.3%
- E 112.7%
Show the worked solution
Answer: B — 31.6%
An index is not a percentage. The base year is 100 by definition, and everything else is measured against that year — so the 2020 figure of 112.7 already means "12.7% above 2015". The question asks about 2020 to 2024, which makes 112.7 the starting value.
(148.3 − 112.7) ÷ 112.7 = 35.6 ÷ 112.7 = 0.316 → 31.6%What each distractor is:
- C — 35.6% is the rise in index points read as a percentage. This is the index-number version of the percentage-point trap, and it is the most-chosen wrong answer.
- A — 24.0% divides by the end value: 35.6 ÷ 148.3. The wrong-base error again, in new clothes.
- D — 48.3% answers a different question — the rise since 2015, which is what the index reads directly. Right arithmetic, wrong period.
- E — 112.7% is a number lifted from the table.
You can read a change from the base year straight off the index (148.3 means +48.3% since 2015). For any other pair of years you must divide. The table is designed so that the easy reading is available and wrong.
Free practice question 6 — the denominator you have to build
| Heading | 2023 | 2024 |
|---|---|---|
| Cohesion | 62.4 | 66.1 |
| Agriculture | 53.8 | 54.2 |
| Research | 12.9 | 14.7 |
| Other headings | 41.9 | 44.0 |
What share of total 2024 commitments did Research represent?
- A 7.5%
- B 8.2%
- C 8.9%
- D 14.7%
- E 27.1%
Show the worked solution
Answer: B — 8.2%
There is no "total" row. You have to build the denominator, and the whole question is really a test of whether you notice that.
66.1 + 54.2 + 14.7 + 44.0 = 179.0 14.7 ÷ 179.0 = 0.082 → 8.2%C — 8.9% is the trap worth remembering: it divides by 179.0 − 14.7 = 164.3, leaving Research out of its own total. A share of a whole includes the part you are measuring. A — 7.5% does the correct calculation on the 2023 column, and D is again a figure copied out of the table.
Adding four numbers under time pressure is where arithmetic slips happen, not in the division. Add the column once, write the total down, and do not re-add it to check — re-check the magnitude of your answer instead. Research is plainly the smallest heading of four, so anything above about 15% is wrong whatever the sum said.
Free practice question 7 — when the shortcut does not work
| Country | Fatalities | Population (m) |
|---|---|---|
| Romania | 1,545 | 19.0 |
| Bulgaria | 531 | 6.4 |
| Sweden | 227 | 10.5 |
| Ireland | 184 | 5.1 |
Which country had the most road fatalities per million inhabitants?
- A Romania
- B Bulgaria
- C Sweden
- D Ireland
- E Cannot be determined
Show the worked solution
Answer: B — Bulgaria, at about 83 per million
Put this next to question 4 deliberately. There, one row dominated on both axes and you could answer by inspection. Here you cannot: Romania has by far the largest numerator, but also by far the largest denominator, so inspection tells you nothing.
Romania 1,545 ÷ 19.0 = 81.3 Bulgaria 531 ÷ 6.4 = 83.0 Sweden 227 ÷ 10.5 = 21.6 Ireland 184 ÷ 5.1 = 36.1A — Romania is the answer you give if you read the largest number in the table and stop, and it is wrong by a whisker: 81.3 against 83.0. That closeness is the point of the question.
Scan for a row that is largest on top and smallest on the bottom. If one exists, answer immediately. If none does — as here — you have to divide, but only for the plausible candidates: Sweden and Ireland are obviously out on inspection, so this is two divisions, not four.
And when two candidates land within a few percent of each other, finish the arithmetic. That is exactly the case where estimation stops being safe.
Free practice question 8 — running a percentage backwards
After an increase of 15%, a budget line stands at €92m. What was it before the increase?
- A €77.0m
- B €78.2m
- C €80.0m
- D €105.8m
- E €107.0m
Show the worked solution
Answer: C — €80.0m
The €92m already contains the increase. Going forwards you would multiply by 1.15; going backwards you divide by it.
92 ÷ 1.15 = 80 check: 80 × 1.15 = 92 ✓B — €78.2m is the trap, and it is a good one: it takes 15% off the 92 (92 × 0.85). That would be right if the question said prices then fell by 15%. It does not — it asks what number, increased by 15%, gives 92. Subtracting 15% of the wrong base overshoots, because 15% of 92 is larger than 15% of 80.
D — €105.8m runs the percentage the wrong way entirely (92 × 1.15), and E is close enough to D to catch a misread.
Reverse-percentage answers are the easiest on the paper to verify: multiply your answer back up and see whether you land on the figure given. If 80 × 1.15 had not returned 92, you would know before moving on. Almost no other question type offers a check this cheap — use it.
Eight questions is a taste. The test is 10 in 20 minutes.
EPSO Turbo is a 6-week programme built around exactly this: 186 questions, each with the step-by-step correction you just read, plus theory sheets, a built-in tutor and timed mock papers from week 5. One payment, no subscription.
Get access — €20The paid programme is currently delivered in French; an English edition is in preparation.
The three traps that cost most of the marks
1. The base of the percentage
More marks are lost here than anywhere else on the paper. "A rose by 40% compared with B" and "B is 40% lower than A" describe two different pairs of numbers. Before dividing, say out loud which value is the reference — the one that follows compared with, than, or of goes on the bottom.
2. Units that shift between columns
Thousands in one column, millions in the next; a rate per 1,000 inhabitants beside an absolute count; euro in the table and a question asked in thousands of euro. The table caption and the column headers carry this information and candidates skip them under time pressure. Read the headers first, before the question — it takes four seconds and it is the cheapest insurance on the paper.
3. Precision you were never asked for
Look at the answer options before you calculate. If they are €89,300m, €111,527m and €124,400m, they are 20% apart and a rough estimate separates them. Computing to the euro is wasted time that you pay for on the next question. Only when two options are close — 41.6% and 42.1% — do you actually need the full calculation.
A two-minute routine that survives the clock
- Seconds 0–10 — read the headers, not the table. Caption, column titles, units. Do not read the data yet; you do not know which of it matters.
- Seconds 10–25 — read the question and name the type. Direct lookup, missing data, percentage change, compound, or ratio. The name tells you what to hunt for.
- Seconds 25–35 — glance at the options. How far apart are they? That decides how precisely you need to work. This is the step everyone skips and it is the one that buys the most time.
- Seconds 35–95 — extract only the two or three numbers you need, note them down (paper or on-screen scratchpad, whichever your session gives you), calculate.
- Seconds 95–110 — check the magnitude, not the arithmetic. Is the answer in a believable region? Wrong-base errors show up as answers that are wildly too large or too small.
- Always answer. If you are past two minutes, take your best estimate, mark it, and move. A wrong answer costs the same as a blank, and the next question may be an easy lookup.
Never let a single hard question eat four minutes. Two questions answered calmly are worth more than one answered perfectly, and the difficulty is not ordered — question 9 may be the easiest on the paper.
How to prepare, over 6 weeks
Numerical reasoning rewards spaced repetition more than almost any other selection test, because what you are building is recognition speed: the reflex that says "this is a cross-multiplication" before you have finished reading the question. Recognition is built in short daily sessions and it decays over long gaps, which is exactly the wrong shape for weekend cramming.
A sequence that works:
- Weeks 1–2 — one family at a time, no timer. Ratios and cross-multiplication first, then estimation, orders of magnitude and compound change. Accuracy only. Speed at this stage builds bad habits.
- Weeks 3–4 — abstract reasoning. It shares the test session and, in many notices, the same combined pass mark, so it cannot be left to the last fortnight.
- Week 5 — mixed review, now timed. Mixing types is what forces the naming reflex; practising one type in a block lets you skip that step without noticing.
- Week 6 — full mock papers, under real conditions. Complete blocks at test pace, with the clock visible. This is where you discover that your two-minute routine is a two-minute-forty routine, and fix it.
Whatever you practise with, insist on corrections that show the reasoning. A key that tells you the answer was B teaches you nothing: you already know you got it wrong. What you need to see is which number should have gone on the bottom of the fraction, and why.
Frequently asked questions
How many questions are in the EPSO numerical reasoning test?
In recent competitions the numerical reasoning block has typically been 10 multiple-choice questions with around 20 minutes to answer them — roughly two minutes per question. The exact number and the time allowed are set in each Notice of Competition, so check yours before you build a pacing plan around any figure.
Is a calculator allowed?
Yes — a basic on-screen calculator is provided in the computer-based test, but you cannot bring your own. It does the four operations and little else, and reaching for it with the mouse costs real seconds, which is why most of your speed still comes from estimation rather than typing.
Can I use paper and a pen during the test?
It depends on how your session is delivered. At an accredited test centre you are given a pen and a few sheets of paper. Under remote proctoring you are not: physical paper and pen are forbidden, and an on-screen scratchpad is provided instead. Your invitation states which applies to you.
This is worth checking early, because it changes how you should practise. Working an intermediate result into an on-screen scratchpad is slower than writing it, and finding that out on the day costs marks.
What is the pass mark for EPSO numerical reasoning?
There is no single permanent pass mark. In many recent notices numerical and abstract reasoning are marked together against one combined threshold, while verbal reasoning has its own; some competitions use the reasoning tests only as a ranking filter. The Notice of Competition for your profile is the only reliable source.
How hard is the EPSO numerical reasoning test?
The arithmetic rarely goes beyond percentages, ratios and division — nothing above secondary-school level. The difficulty is the clock: about two minutes to read a dense table, work out which cells you need, and avoid the wrong-base and percentage-point traps. Candidates fail on method and pacing far more often than on calculation.
Is it the same test for AD and AST competitions?
The question format is the same — a table or chart, a question, five options — but the number of questions, the time allowed and the pass marks differ between AD, AST and AST-SC profiles and between competition years. The skills you train are identical; only the pacing target changes.
Which language is the test taken in?
In most recent competitions the reasoning tests are taken in your language 1. Numerical reasoning is the least language-dependent of the three: once you can read the column headings, the work is arithmetic. Your Notice of Competition states the language regime that applies to you.
How long should I practise before the test?
Six to eight weeks of short daily sessions, rather than a few long weekends. Speed here comes from pattern recognition, and recognition is built by spacing practice out — cramming produces candidates who can solve every question type slowly.