Numerical Reasoning Test: Formats and Traps

A numerical reasoning test asks you to answer questions from tables, charts and short word problems under time pressure, using percentages, ratios, averages, unit rates and growth. Employers in consulting, banking, logistics and the public sector use them early in the process. The usual format is 12 to 30 multiple-choice items in 10 to 36 minutes, though some tests record your time instead of limiting it. This guide covers every format seen in real 2026 tests from SHL, TestGorilla, Cappfinity and Applied, the traps that made them hard, and 12 original practice questions with worked answers.

Key facts: the formats compared

What numerical reasoning tests look like

1. Word problems (multiple choice)

A few sentences, one question, four or five options. Typical topics: percentage of a total, reverse percentages (a price after a rise), two unknowns ("A costs £x more than B"), work-rate ("12 people take 6 days…"), proportional scaling, and time puzzles with fast and slow clocks. They are quick if you set up the equation first.

2. Table and chart items

The classic format: a table or chart, then a question about it. Seen in every vendor. Variations: - Labelled charts with values printed on the bars (Cappfinity). - Unlabelled charts where you read heights against gridlines (TestGorilla), with a zoom icon for detail. - Interactive charts with hover tooltips that show the exact value only when you point at it (Applied, Cappfinity). The axis may be enough for an estimate, but close options need the tooltip. - Figures below the fold: "Please use the 4 figures below… You may need to scroll down to see them all." Tables that scroll sideways hide their last column. - One data set, two questions: the same table returns for a second item.

3. Data literacy

Instead of calculating, you pick the right method or judge a claim: which formula gives next year's rent after a rise, how to recover a pre-sale price, which column needs cleaning, which chart type fits, whether a summary of a truncated-axis chart is fair, which statements a table supports ("Select all that apply"). TestGorilla ran this as a separate 10-minute test.

4. Step and process calculations

Multi-step items: a day-rate with a discount on some days only; a salary with a percentage raise plus a fixed amount each year; a budget remaining this month from an annual figure; a process loop ("Step 1: multiply by 3. Step 2: if under 40, go back to Step 1…") where you test each starting value.

5. Data sufficiency and "information needed"

Five fixed options ("Statement 1 alone is sufficient…", "Each statement alone is sufficient", "Both together are sufficient but neither alone", "Not sufficient together") or a list of facts from which you choose the ones needed. Seen in Applied's test, with the option order shuffled per item.

6. Error checking

Compare two columns and count mismatches, or find the one cell in a table whose calculation is wrong and caused the other errors. Seen in Applied's numerical test. See error-checking tests.

7. Number series

"What comes next?" sequences inside a numerical test (TestGorilla). See number series.

Number formatting traps

• Decimal comma vs decimal point. In several 2026 tests, chart labels or tooltips used continental formatting ("2.418,6" means two thousand four hundred and eighteen point six; "27,35" means twenty-seven point three five), while the answer options used UK or US formatting. This is probably locale-driven. Before calculating, rewrite every figure in one format.
• Inconsistent separators in one item. One SHL item printed most options with a decimal comma and one with a point. Read each option for its value, not its punctuation.
• Units in thousands or millions. A header "(£k)" or "(£000s)" turns a difference of 240 into £240,000. The unit-slip distractor (one zero short) is almost always offered.
• Answers written in words. Some items give options like "three million, four hundred thousand". Convert to digits before comparing.
• Typed answers. "Insert your answer in £ to two decimal places" means 52.10, not 52.1 or £52.10 with a symbol unless asked.

Misleading axes

• Truncated y-axis. If the axis starts at 80,000 instead of zero, a bar that looks half as tall may be 3% lower. Always read the axis minimum before judging "doubled", "halved" or "three times".
• Mislabelled gridlines. One TestGorilla chart had a gridline label that broke the axis pattern. Work out the step from two clean labels.
• Two axes. Combined bar and line charts may use a second axis on the right. Check which series belongs to which axis.
• Percentage-change charts. A chart of "change on the previous month (%)" is not a chart of levels. To get levels you start from a known month and work forwards and backwards.

12 original practice questions with worked answers

1. Word problem, percentage of a total. Three call centres receive 1,200, 1,800 and 3,000 calls a day. Each centre's quality team reviews 60 calls a day. What percentage of all calls is reviewed? 1.00% / 3.00% / 3.33% / 3.44% / 5.00%

Answer: 3.00%. Calls reviewed: 3 × 60 = 180. Total calls: 6,000. 180 ÷ 6,000 = 3%. Averaging the three centres' rates (5%, 3.33%, 2%) gives the trap 3.44%; using one team only gives 1%.

2. Table, reverse percentages. This year's revenue and growth on last year: Region A €6.12m (+2%), Region B €4.83m (+15%), Region C €3.50m (−12.5%). By how much did Region A's revenue exceed Region C's last year? €2.62m / €2.00m / €1.29m / €2.15m

Answer: €2.00m. Last year A = 6.12 ÷ 1.02 = 6.00; C = 3.50 ÷ 0.875 = 4.00. Difference 2.00. Using this year's figures gives 2.62.

3. Decimal comma in tooltips. Hovering over a cost chart shows January "1.084,6", February "947,2" and March "1.102,0" (€). What is the first-quarter total? €949.39 / €3,133.80 / €2,186.60 / €3,138.80

Answer: €3,133.80. Rewrite as 1,084.6 + 947.2 + 1,102.0 = 3,133.8. Reading "1.084,6" and "1.102,0" as 1.0846 and 1.102 gives the trap 949.39; €2,186.60 leaves out February.

4. Truncated axis. A column chart of weekly visitors has a y-axis from 480 to 530. Week 1 is 500 and week 10 is 520, so the week 10 bar looks twice as tall. Which statement is correct? Visitors doubled / Visitors rose by 4% / Visitors rose by 40% / Visitors rose by 20 per cent

Answer: Visitors rose by 4%. 520 ÷ 500 = 1.04. The bar heights above 480 (20 and 40) create the doubling illusion.

5. Hours and minutes. Plant A ran 5 h 40 min and made 13,430 units; Plant B ran 6 h 00 min and made 14,520 units; Plant C ran 4 h 15 min and made 10,370 units. Which plant was least efficient? Plant A / Plant B / Plant C / A and C are equal

Answer: Plant A. Convert minutes to hours: 5 h 40 min = 5.667 h, so A = 13,430 ÷ 5.667 = 2,370 units per hour. B = 14,520 ÷ 6 = 2,420. C: 4 h 15 min = 4.25 h, so 10,370 ÷ 4.25 = 2,440. Misreading 5 h 40 as 5.40 h and 4 h 15 as 4.15 h makes A look like 2,487 and C like 2,499, and points to B, which is the trap.

6. Data sufficiency. A team has 40 people, each either full-time or part-time. What percentage works part-time? (1) There are 12 more full-time than part-time staff. (2) Half the team are women, and 35% of the women work part-time.

Answer: Statement 1 alone is sufficient, but statement 2 alone is not. From (1): F + P = 40 and F − P = 12, so P = 14, which is 35%. Statement (2) says nothing about the men, so it cannot fix the total. The matching 35% in (2) is a lure.

7. Compound growth. A fund holds €12,500 and grows 4% a year for three years. What is it worth at the end, to the nearest euro? €14,000 / €14,061 / €13,520 / €14,623

Answer: €14,061. 12,500 × 1.04³ = 12,500 × 1.124864 = 14,060.80. Adding 3 × 4% = 12% gives the simple-interest trap €14,000.

8. Data literacy, formula choice. Office rent is €5,000 a month. From 1 April it rises by 3%. Which calculation gives the rent for the calendar year? 5,000 × 1.03 × 12 / 5,000 × 3 + 5,000 × 1.03 × 9 / 5,000 × 0.03 × 12 / 5,000 × 12 + 0.03

Answer: 5,000 × 3 + 5,000 × 1.03 × 9 = 15,000 + 46,350 = €61,350. January to March are at the old rate. The first option (€61,800) assumes the rise starts in January.

9. Multi-figure item with a red herring. Figure 1 (a line chart) shows export values in € million for Country X and Country Y; hovering over 2025 shows X "38,6" and Y "12,4". Figure 2, below the fold, notes "Exchange rate: 1 € = 1.08 US$". What were the two countries' combined exports in 2025, in € million? 51.0 / 52.0 / 55.1 / 50.1 / None of these

Answer: 51.0. Both series are already in euros: 38.6 + 12.4 = 51.0. The exchange rate is a distractor: converting Y alone gives 52.0, converting the total gives 55.1.

10. Number series. 5, 8, 14, 26, 50, ? 74 / 86 / 98 / 100

Answer: 98. Each term is double the previous minus 2: 50 × 2 − 2 = 98. The gaps (3, 6, 12, 24, 48) also double.

11. Error checking. Compare the two columns and count the rows that do not match.

Row  Column A     Column B
1    KX-4471-B    KX-4471-B
2    PL-0938-A    PL-0938-A
3    MT-5620-C    MT-5602-C
4    RD-1184-D    RD-1184-D
5    ZN-7719-A    ZN-7719-A
6    HB-3305-E    HB-3305-F
7    QC-8862-B    QC-8862-B
8    WF-2047-C    WF-2047-C

1 / 2 / 3 / 4

Answer: 2. Row 3 swaps two digits (5620 vs 5602) and row 6 changes the final letter (E vs F). Check digit groups from left to right with a pointer (cursor or finger), one row at a time.

12. Process loop. Step 1: multiply by 3. Step 2: if the result is under 40, return to Step 1; otherwise go to Step 3. Step 3: subtract 10. Step 4: divide by 2. Which starting value gives the largest result? 1 / 2 / 3 / 4 / 5

Answer: 4. Run each value: 1 → 3 → 9 → 27 → 81 → 35.5; 2 → 6 → 18 → 54 → 22; 3 → 9 → 27 → 81 → 35.5; 4 → 12 → 36 → 108 → 49; 5 → 15 → 45 → 17.5. The largest starting number (5) is the trap: it leaves the loop earliest at a low value.

Traps that made real items hard

• Conditions that apply to part of the data: a discount on the first five days only, a rebate with a start date and a minimum value, a price change after the forecast was made.
• Wrong row or column: department vs company budget, forecast vs actual, last year vs this year.
• Digit-permuted options: 47.15% next to 41.75%; the same digits in a different order. Compare digit by digit.
• Ratio order and off-by-one: both 1:2 and 2:1 are offered; the 7th term is offered when the 8th is asked.
• Reverse vs forward percentages: after a 15% rise, divide by 1.15; do not multiply by 0.85.
• Red herrings: an exchange rate when everything is already in one currency, a revenue chart in a cost question, a "hover" hint on a text-only item.
• Near-ties: two values within 1% of each other; estimate first, but calculate exactly before you choose.
• Flawed or ambiguous items: some real tests contain an item whose options do not match a literal reading. If no option fits after a careful second read, choose the closest and move on.

Speed techniques

1. Read the question before the data. Then look only for the figures you need.
2. Write the unit next to every number you note down (£k, %, hours).
3. Estimate, then calculate. A rough answer (about 40%) rules out two or three options and protects you from calculator slips.
4. Use fraction anchors: 1/8 = 12.5%, 1/6 = 16.7%, 1/3 = 33.3%, 3/8 = 37.5%, 5/8 = 62.5%, 7/8 = 87.5%.
5. Multipliers for change: +15% is × 1.15, −12.5% is × 0.875; two changes multiply.
6. Time boxes. Divide the time by the items (10 min / 12 items ≈ 50 s; 30 min / 18 items = 100 s). Skip or guess at 1.5 × that.
7. Bank quick items. In tests with a Previous button (Applied), answer the one-step items first; in forward-only tests, never leave an item blank.
8. Hover and scroll first. For interactive charts, collect every tooltip value you need and scroll to the end of the panel before calculating.

How to prepare

Two weeks, 30 minutes a day, or one intensive week at an hour a day.

Setup. Use a laptop or desktop with a large screen: hover tooltips, zoom views and side-scrolling tables are hard on a phone. Keep a physical calculator if allowed, and pen and paper. Close other windows and switch off translation extensions, which can change numbers on the page.

Sources

• SHL, Numerical Reasoning example questions
• TestGorilla test library, Numerical Reasoning test (question families, 10 min)
• AssessmentDay, Cappfinity-style numerical tests (time recorded; pick, insert and rank styles)

Written by

Adrian, Specialist

MSc Psychology, 7 years of test taking.

Last updated: Oct 1, 2026