Logical Reasoning Tests: Formats, Worked Questions & Practice Guide

A logical reasoning test is a timed assessment that measures how well you identify patterns, apply rules and draw valid conclusions from information you have never seen before. The defining feature is that it is designed to be knowledge-free: unlike a technical test or a subject exam, you are not expected to bring any domain expertise. Everything needed to answer each question is contained in the question itself.

That is precisely why employers like these tests. A graduate applicant pool contains candidates from dozens of different degree subjects, universities and countries, and grades are not directly comparable across them. A logical reasoning test produces a single number on a single scale for every candidate, which is why it usually sits at the very start of a recruitment process — immediately after the online application form and before anyone reads your CV in detail.

Confusingly, "logical reasoning test" is not one test. It is a label that employers, recruiters and test publishers use loosely to cover several different formats. When an invitation email says you will sit a "logical reasoning test", it could mean any of the following:

• Deductive reasoning — applying stated rules or premises to reach a conclusion that must be true. Often word-based: syllogisms, seating arrangements, scheduling puzzles. See our full deductive reasoning guide.
• Inductive reasoning — inferring the underlying rule from a set of examples, then extending it. Usually shape-based sequences. See our inductive reasoning guide.
• Abstract reasoning — identifying relationships between non-verbal, non-numerical figures. Frequently used interchangeably with inductive reasoning. See our abstract reasoning guide.
• Diagrammatic reasoning — following inputs through a sequence of process boxes or operators to determine the output. See our diagrammatic reasoning guide.

What all four share is the underlying construct psychologists call fluid intelligence — your capacity to solve novel problems under time pressure. Research in occupational psychology has consistently found general cognitive ability to be among the strongest single predictors of job performance across role types, which is the evidence base employers point to when they use these tests as a sift.

The table below sets out how the four formats typically differ. Exact question counts and time limits vary considerably by publisher, test version and role level, so treat these as commonly reported ranges rather than fixed specifications — always rely on the figures given in your own invitation.

Deductive vs inductive — the distinction that matters

These two are routinely confused, including by recruiters, but the mental operation required is genuinely different. In deductive reasoning you are given the rule and must apply it rigorously — the correct answer follows necessarily from the premises, and anything not stated cannot be assumed. In inductive reasoning the rule is hidden and you must reverse-engineer it from examples, which means the answer is the best-supported inference rather than a logical certainty.

Practically, this changes your strategy. On deductive items the error to guard against is importing real-world knowledge or plausible assumptions the passage does not actually contain. On inductive items the error is locking onto the first pattern you spot and failing to test it against every example in the sequence.

Across the four formats there are roughly six recurring question architectures. Recognising which one you are looking at within the first few seconds is itself a learnable skill, and it is the single biggest source of time savings available to you.

1. Syllogisms and conditional statements

You are given two or more premises ("All project managers attend the Monday briefing. Priya does not attend the Monday briefing.") and must evaluate a conclusion. The key discipline is treating the premises as the entire universe of facts. The most common trap is the invalid reversal: "All A are B" does not license "All B are A".

2. Arrangement and constraint puzzles

Five people in a queue, four shifts across a week, three offices on two floors — with a list of constraints. These reward sketching. Draw the positions as boxes, enter every certainty first, then use the constraints to eliminate. Attempting them purely in your head is where candidates lose minutes.

3. Shape sequences

A row of four or five figures changes according to one or more rules; you select the next figure. Rules almost always come from a small, finite set: number of elements, shape type, shading, colour, rotation, position, size, reflection and the addition or removal of a feature. Checking those dimensions in a fixed order is far faster than staring at the sequence as a whole.

4. Matrices

A three-by-three grid with the bottom-right cell missing. Patterns typically run along rows, down columns, or both simultaneously; some use addition or subtraction of elements between cells. This is the architecture made famous by Raven's Progressive Matrices, and the method is to establish the row rule first, then confirm it against the column rule before looking at the options.

5. Odd-one-out

Five figures share a property; one does not. The efficient approach is to find what the majority have in common rather than searching for what makes one figure unusual — there are usually several superficial ways any single figure differs, and only one that four of the five share.

6. Input/output flowcharts

A set of symbols is defined as operators (one rotates, one swaps positions, one changes shading), and you trace an input through a chain of them. Write the intermediate state down after each operator. Candidates who try to hold three transformations in working memory get the direction of one of them wrong far more often than they expect.

7. Letter and number series

Sequences such as B, D, G, K... where the step size itself changes. Convert letters to their alphabet positions (B = 2, D = 4, G = 7, K = 11), then examine the gaps: 2, 3, 4. The next gap is 5, giving position 16 — P. This conversion step takes seconds and turns an opaque letter puzzle into straightforward arithmetic, with none of the data interpretation you would meet in a numerical reasoning test.

The examples below are written out in words rather than shown as images, which makes the reasoning easier to follow. Work through each one yourself before reading the solution.

Example 1 — Deductive (syllogism)

Premises: All of the firm's auditors have completed the ethics module. Some of the people who have completed the ethics module work in the Leeds office. No one in the Leeds office is a partner.

Conclusion to evaluate: "Some auditors work in the Leeds office." Is this true, false, or impossible to say?

Answer: impossible to say. The second premise tells us that some ethics-module completers are in Leeds, but the set of completers is larger than the set of auditors — it includes auditors plus potentially many other staff. The Leeds completers could consist entirely of non-auditors. Nothing in the premises forces any auditor into Leeds, and nothing forbids it either.

The trap: the overlap feels intuitive because both statements mention the ethics module. Formally, "All A are B" plus "Some B are C" never yields "Some A are C". If you find yourself constructing a plausible story to justify an answer, that is the signal to pick "cannot say".

Example 2 — Inductive (shape sequence)

Sequence:

• Figure 1: one black triangle, pointing up, in the top-left of the box
• Figure 2: two black triangles, pointing up, in the top-right
• Figure 3: three white triangles, pointing up, in the bottom-right
• Figure 4: four white triangles, pointing up, in the bottom-left

Question: what is Figure 5?

Answer: five black triangles, pointing up, in the top-left. Run the dimensions in order. Count: 1, 2, 3, 4 → 5. Shading: black, black, white, white → the cycle has a period of four, so Figure 5 returns to black. Position: top-left → top-right → bottom-right → bottom-left, a clockwise rotation around the corners, so Figure 5 returns to the top-left. Orientation: unchanged throughout, so it stays pointing up.

Why the method matters: three rules are running simultaneously with different periods. Scanning count, shading, position and orientation as a fixed checklist catches all three. Looking at the figures holistically, most candidates spot the count and miss the shading cycle.

Example 3 — Diagrammatic (input/output)

Operators: a circle operator rotates the input 90° clockwise. A square operator swaps the two halves of the input left-to-right. A triangle operator inverts the shading of every element.

Input: a tile with a black dot on the left half and a white dot on the right half. It passes through: square, then triangle, then square.

Answer: a white dot on the left and a black dot on the right. Track the state after every operator, writing the left and right halves as two letters:

• Start: BW — black left, white right.
• After the square (swap halves): WB — white left, black right.
• After the triangle (invert shading): BW — the white left becomes black, the black right becomes white.
• After the second square (swap halves again): WB — white left, black right.

Notice that the two swaps cancel each other out, so the net effect of the whole chain is the shading inversion alone — the output is the inverse of the input, not the input itself. Spotting cancelling operators can save time on longer chains, but only once you have tracked the states carefully enough to be sure they really do cancel. The two most common failure modes here are applying transformations in the wrong order and silently skipping a step; writing the two-letter state after each operator takes about three seconds and eliminates both.

Almost no employer looks at your raw score. What they see is a percentile, and understanding how one becomes the other explains most of what candidates find confusing about results.

From raw score to percentile

Your raw score is the number of items you answered correctly. The publisher then compares that raw score against a norm group — a reference sample of people who have previously taken the same test — and reports where you fall in that distribution. A percentile of 70 means you scored higher than approximately 70% of the norm group.

The norm group choice therefore matters enormously. The same raw score can convert to very different percentiles depending on whether it is benchmarked against the general working population, graduate applicants, or a specialist group such as finance graduates. This is why a score that looked strong for one employer can be unremarkable for another, and why percentile reports usually name the comparison group used.

What counts as a pass

Most employers do not publish their thresholds. Based on widely reported candidate experience, graduate schemes commonly sift somewhere around the 50th to 70th percentile, with the most competitive employers reported to set the bar higher. The practical implication is straightforward: aim to be comfortably above the 70th percentile in practice rather than targeting a specific number, because you cannot know the threshold in advance. Our guide to how SHL assessments work covers how results are reported and used in more detail.

Adaptive and computer-adaptive tests

Many modern logical reasoning tests are adaptive: the difficulty of the next item is selected based on whether you answered the previous one correctly. Three consequences follow. First, two candidates sitting "the same" test may see entirely different questions. Second, the test can estimate your ability level from fewer items, so adaptive tests are often shorter. Third, you usually cannot go back and change an earlier answer, because the item you were served next depended on it.

Accuracy versus speed

Many tests report accuracy and speed as separate indicators. Where unanswered items count as incorrect and there is no negative marking — the usual arrangement — guessing on an item you cannot solve is strictly better than leaving it blank. On a five-option item that is a 20% expected return for a few seconds of effort. Check your instruction screen, since a small number of tests do apply a correction for guessing.

Logical reasoning tests are produced by a relatively small number of assessment publishers and licensed to employers. Identifying the publisher behind your test tells you a great deal about the format you will face.

Which employers use them

The pattern is consistent across sectors. Investment banks, professional services firms and management consultancies use logical reasoning alongside numerical tests for graduate intake. Technology employers use diagrammatic and abstract formats for engineering and analyst roles. Engineering, rail, defence and utilities employers favour diagrammatic and spatial reasoning formats for technical positions. Public sector and civil service processes typically combine reasoning tests with situational judgement.

Where critical thinking matters more than pattern recognition — particularly in law — employers often use the Watson Glaser critical thinking appraisal instead, which assesses inference, assumption recognition and argument evaluation rather than abstract pattern spotting. For a broader map of how these tests fit together, see our overviews of aptitude tests and psychometric tests.

Logical reasoning tests respond well to short, focused preparation. Most of the available gain comes from familiarity with question architectures and from fixing your timing strategy — both achievable in a week of deliberate practice.

The six techniques worth the most marks

• Scan dimensions in a fixed order. For any figure-based item, check count, shape, shading, rotation and position in the same sequence every time. A systematic scan beats intuition because it catches the second and third rules running in parallel.
• Eliminate rather than construct. You do not need to derive the correct answer — you need to rule out four wrong ones. Identifying a single disqualifying feature in an option is usually much faster than fully solving the item.
• Never import outside knowledge on deductive items. If a conclusion requires a fact the passage does not state, however obviously true in the real world, it is not supported. "Cannot say" is the correct answer far more often than candidates expect.
• Write down intermediate states. On diagrammatic and arrangement items, three seconds of notation prevents the most common error class in the entire test.
• Budget per item and enforce it. One item solved perfectly in four minutes costs you three items you would have got right. The marginal value of your last 30 seconds on a hard item is almost always negative.
• Always answer every question. Where there is no guessing penalty, an unanswered item is a guaranteed zero and a guess is a 20–25% chance. Over a handful of items that difference can move your percentile meaningfully.