Free numerical reasoning practice test
A numerical reasoning practice test is a short, unscored rehearsal of the data-interpretation questions used in graduate and professional selection: read a table or chart, choose the right operation, and answer under time pressure. Practice reliably improves familiarity and speed. It does not raise the underlying ability the full-length test is built to measure.
Eight questions, an advisory eight-minute clock, scored the moment you press the button. There is no account, no email field and no paywall. Everything — the items, the key, the working — is already in this page, and the scoring happens in your browser, so nothing you do here is sent to AssessAll or stored anywhere.
One thing here that you will not find on other free practice tests: the score comes back with the margin of error around it. On eight items that margin is enormous, and printing the number without it is the reason so many people walk away from a short practice test with a badly wrong idea of where they stand.
Founder of AssessAll and of Bodhih Training Solutions, a corporate training company in Bangalore. Works on assessment design, scoring and reporting across hiring, L&D and certification programmes.
Last reviewed
The test: 8 questions, 8 minutes
Answer as many as you like and press Score my answers. Unanswered items count as wrong, which is how the real form treats them too. You can restart as often as you want.
- Question 1 of 8Percentage change
Support tickets resolved, two quarters
Team Q1 Q2 Billing 1,240 1,426 Onboarding 880 946 Technical 2,100 2,037 By what percentage did the Billing team's resolved tickets change from Q1 to Q2?
- Question 2 of 8Ratio and share
Headcount split across three sites
A company employs 1,155 people across three sites. The headcount is split between Manchester, Leeds and Sheffield in the ratio 5 : 4 : 2.
How many people work at the Leeds site?
- Question 3 of 8Reading a rate from a table
Cost per unit across suppliers
Supplier Units Total cost (£) Aldren 3,400 27,880 Bracken 2,750 22,825 Corley 5,100 40,290 Which supplier has the lowest cost per unit, and what is it?
- Question 4 of 8Percentage of a percentage
Conversion through a two-stage funnel
Of 8,400 applicants, 25% are invited to an assessment. Of those invited, 36% are invited again to a final interview.
How many applicants reach the final interview?
- Question 5 of 8Reverse percentage
A price quoted after tax
An invoice total of £4,260 includes VAT charged at 20% on the pre-tax amount.
What was the amount before VAT?
- Question 6 of 8Weighted average
Average order value across two channels
Channel Orders Average order value (€) Direct 1,800 62.00 Partner 700 94.00 What is the average order value across both channels combined?
- Question 7 of 8Rate, time and scaling
Throughput on a processing line
Four machines running together process 1,920 components in 6 hours. All machines work at the same constant rate.
How long would seven machines take to process 5,600 components?
- Question 8 of 8Reading what the data does not say
Year-on-year revenue by region
Region 2024 (€m) 2025 (€m) North 48.2 51.6 South 39.7 38.1 East 22.4 29.0 Which of these statements is fully supported by the table alone?
How much is an eight-question score actually worth?
Very little on its own, and the arithmetic says so precisely enough that there is no need to hedge. Every score out of 8 carries a confidence interval, and on eight items those intervals overlap so heavily that most scores are statistically indistinguishable from most other scores. Here is the whole table, computed with the Wilson score method rather than the textbook formula, which misbehaves badly at this sample size:
| Score | Percentage | 95% confidence interval | Width |
|---|---|---|---|
| 0 / 8 | 0.0% | 0.0% – 32.4% | 32 pts |
| 1 / 8 | 12.5% | 2.2% – 47.1% | 45 pts |
| 2 / 8 | 25.0% | 7.1% – 59.1% | 52 pts |
| 3 / 8 | 37.5% | 13.7% – 69.4% | 56 pts |
| 4 / 8 | 50.0% | 21.5% – 78.5% | 57 pts |
| 5 / 8 | 62.5% | 30.6% – 86.3% | 56 pts |
| 6 / 8 | 75.0% | 40.9% – 92.9% | 52 pts |
| 7 / 8 | 87.5% | 52.9% – 97.8% | 45 pts |
| 8 / 8 | 100.0% | 67.6% – 100.0% | 32 pts |
Read the middle of that table and the problem is obvious. Someone who scores 5 out of 8 has a true rate somewhere between 30.6% and 86.3%. Someone who scores 7 out of 8 has a true rate somewhere between 52.9% and 97.8%. Those two ranges overlap across more than thirty percentage points, so the two people cannot be reliably separated by this test — even though one of them scored two items higher and would, on any other practice site, be told they had done substantially better.
There is a second way to see the same thing. The Spearman–Brown prophecy formula predicts what happens to a test's reliability when you change its length. A full-length commercial numerical form of 25 items with a published reliability of 0.85 — a typical figure for instruments in this family, and quoted here as an illustration rather than as an AssessAll measurement — becomes, at 8 items, 0.64. And that assumes the eight items are a parallel sample of the full form, which flatters the short version: a hand-picked eight is usually less internally consistent than a random eight, not more.
None of this makes practice pointless. It makes the score close to pointless while leaving the practice itself valuable, and those are different things. What eight items can give you is exposure to the item formats, a feel for the clock, and — because the working is printed for every option — a diagnosis of which operations you are actually getting wrong. That last one is the part worth having, and it does not depend on the score being reliable at all.
What AssessAll actually runs, and what it does not
AssessAll is an assessment platform, not a test-prep site, and this page is a rehearsal rather than a sample of the product. The full-length instrument this teaser is modelled on is Numerical Reasoning — Graduate: 25 items in 22 minutes, multiple choice, scored right/wrong against a computed key with no negative marking.
What AssessAll does not publish is a reliability coefficient, a norm table, a benchmark or a percentile for any of its own instruments. That is not an oversight. The platform holds an internal research-readiness gate — a minimum of 400 completed sittings on the single assessment being reported, from at least ten distinct organisations, with no organisation contributing more than a quarter of the sample — and no instrument has yet cleared it. A page arguing that eight items is not enough evidence cannot itself publish an under-evidenced number, so it does not.
Where these questions came from
Every item on this page was written for this page. None of them is drawn from any AssessAll form, from the live item bank, or from anything a candidate might sit. Publishing operational items would degrade the instrument for every organisation using it, which is why no reputable publisher circulates its item bank — and it is the honest answer to “why won't you show me the real questions”, which most sites in this lane simply avoid being asked.
What this page is not
- This is not a selection test and no result from it is used, stored or seen by anyone. The scoring runs entirely in your browser.
- Eight items cannot produce a reliable individual score, and this page does not pretend otherwise — it reports the margin of error alongside the number, which is the whole reason the margin is on the page.
- No percentile is offered, because a percentile is a property of a comparison group and there is no comparison group here. A percentile computed against nobody is a decoration.
- These items were written for this page. They are not taken from any AssessAll form, and a score here predicts nothing about a score on one.
- AssessAll publishes no reliability coefficient, norm or benchmark for any of its own instruments, because none has yet passed its internal research-readiness gate. The 0.85 used in the arithmetic below is a typical published figure for full-length commercial forms in this family, quoted as an illustration and labelled as one.
Common questions about numerical reasoning practice
Is this numerical reasoning practice test really free, with no signup?
Yes. There is no account, no email field and no paywall. The eight items, the key, the full working and the score all load with the page and are scored in your browser. Nothing you do on this page is transmitted to AssessAll or stored anywhere, which is also why your result disappears when you close the tab.
How reliable is an 8-question aptitude test score?
Much less reliable than the number makes it look. If a full 25-item form had a reliability of 0.85, the Spearman–Brown formula puts an 8-item version of it at about 0.64 — and 6 out of 8 carries a 95% confidence interval running from 40.9% to 92.9%, a band 52 percentage points wide. That interval is not a criticism of any particular short test; it is arithmetic that applies to every one of them, including this one. It is why no selection decision should ever rest on eight items, and why this page prints the interval next to the score rather than the score alone.
What is a good score on a numerical reasoning test?
There is no universal answer, because a raw score means nothing without knowing the difficulty of the form, the time allowed and the comparison group. On a real graduate form, employers typically set a cut score against a norm group or against a criterion, not against a fixed percentage. On this page the honest answer is narrower: eight items cannot tell you whether your score is good, only roughly where a much wider band around it sits.
Does practising numerical reasoning tests actually improve your score?
It improves the part of the score that comes from familiarity — knowing the item formats, not re-deriving reverse percentages under time pressure, and not losing thirty seconds working out what is being asked. Published work on test coaching finds real but modest gains that are largest on the first few exposures and flatten quickly. Practice does not raise the underlying reasoning ability the test is built to measure, and any site promising that it does is selling something.
Are these the same questions AssessAll uses in its real tests?
No, and deliberately so. Every item on this page was written for this page. Publishing live items would degrade the instrument for every organisation using it, which is the same reason no reputable test publisher circulates its operational item bank. The trade-off is that these items are representative of the format rather than calibrated against it.
Is a timer required?
No. The eight-minute clock is advisory and you can ignore it or let it run out — the test does not stop and nothing is submitted when it reaches zero. It is there because time pressure is a real component of most graduate numerical forms, and practising untimed teaches you a different task from the one you will sit.
Read next
- What a numerical reasoning test measures
The full-length picture: what the construct is, how a speeded and a power form of it differ, and five further worked items.
- Sample size & reliability calculator
Run the confidence-interval arithmetic on this page for any test length, not just eight items.
- Standard error of measurement
The quantity that decides how wide the band around any single score has to be.
- Verbal reasoning: reasoning or English?
The sibling construct, and the one whose name covers two different instruments.
- What is a psychometric assessment?
Where reasoning tests sit among the other instrument families.
Sources
- Wilson, E. B. (1927), Journal of the American Statistical Association 22(158), 209–212 — The score interval used for the confidence band on the result.
- Agresti, A. & Coull, B. (1998), The American Statistician 52(2), 119–126 — Why the textbook Wald interval is the wrong tool at n = 8.
- Brown, W. (1910) / Spearman, C. (1910), British Journal of Psychology 3, 271–295 / 296–322 — The prophecy formula, run backwards to estimate the reliability of a shortened form.