Abstract Reasoning Test: Pattern Rules and Preparation

Abstract reasoning tests can feel strange because they use shapes and patterns instead of ordinary workplace information. But the questions are not random. They are built from repeatable visual rules. This guide explains what abstract reasoning tests measure, the common pattern families, and how to practise so the rules become easier to recognize under time pressure.

A matrix item is rarely one puzzle. It is usually two or three independent rules stacked in the same frame, each governing a separate attribute, and capable people stall because they read whole cells instead of attributes. Stare at a cell holding four triangles, a shaded border and an arrow, and your eye takes in all three at once, finds no pattern, and offers only a hunch that the answer should resemble the cell above. Split the same grid into three passes, counting triangles on the first, reading shading on the second, arrow direction on the third, and each pass resolves into something trivial: add one, alternate, rotate a quarter turn. The difficulty of an abstract reasoning matrices test sits in the superposition of simple rules, not in the rules themselves, which is also why wrong options feel plausible: each obeys every rule but one.

Practising for a real assessment? The Abstract & Inductive Reasoning Practice Pack runs these question types at full length — 5 timed tests, 150 questions, a worked explanation on every one.

The formats inside the abstract reasoning label

One label covers several mechanically different tasks. They share a look, so candidates assume a single skill, then find a score on one does not transfer to the next.

3 x 3 matrix completion

Eight cells sit in a grid with the ninth, usually bottom right, missing. Rules run along rows, along columns and occasionally along a diagonal, and the defining feature is that several run at once over different attributes: element count across rows, shading down columns, orientation along a diagonal, none interfering with another.

Attribute isolation is what the format rewards. List what varies: count, shape identity, size, shading, rotation, position within the cell, line thickness. Freeze all but one attribute and read the grid twice, across and down, asking only what happens to that one. Write the rule, then take the next. Three easy rules found separately beat one impossible rule guessed whole.

The rules come from a small vocabulary: progression in a count, constancy along an axis, alternation, rotation by a fixed angle, distribution of three values so each row and column carries each value once, and superimposition, where the third cell in a row combines the first two, sometimes with shared elements cancelling.

Sequence and series completion

Four or five figures run in a line and you supply the next. Mechanically this is a matrix with one axis rather than two: a single progression instead of interacting row and column rules. Simpler to unpick, but less forgiving, because a matrix confirms a rule twice, across and down, whereas a sequence gives one reading and no check.

Odd one out

Five or six figures, one of which does not belong. No progression is involved, which inverts the work: rather than tracking how an attribute changes, you test candidate invariants, such as every figure having an even number of sides, or exactly one curved edge. Well built items support two rival groupings, so a property isolating one figure means little until you have checked whether another isolates a different one.

Analogy pairs

A is to B as C is to what. The transformation between the first pair is the rule and you apply it to C. Mechanically, one example carries the whole rule rather than several steps confirming it, so an ambiguous first pair leaves several defensible transformations, and the options reward the one holding for every attribute, not the obvious one.

Set membership

Two governed groups are shown and you decide which a new figure joins, or which joins neither. Each group carries its own rule, so you test membership twice rather than follow a progression, and a figure can satisfy part of both rules and belong to neither.

Across all five, distractors are built the same way: from rules misapplied in one direction. A row rule read down a column, the right count with shading a step behind, a rotation the right way through the wrong angle. That is why an option looking right proves nothing, and why the reliable close is checking every rule against every surviving option.

Scoring and timing, and what they mean for strategy

Two scoring models are in circulation and they require opposite behaviour. Under raw scoring your mark is the number of correct answers, and a blank scores exactly what a wrong answer scores, so leaving anything unanswered throws away a free chance. Under formula scoring a fraction of a mark is deducted for each wrong answer and nothing is deducted for an unanswered one, which removes the expected gain from random guessing. Where a wrong answer costs a fraction of a mark and an unanswered one costs nothing, filling in the last few boxes blind is a net loss rather than a free chance.

Timing models vary just as much. A fixed question count against a fixed clock makes pacing a division problem and every item reachable. An indefinite bank against a clock means nobody finishes, the total correct is the measure, and four minutes on one stubborn matrix is expensive in a way it never feels at the time. Per-section timers stop you borrowing minutes from the sequence block to rescue the matrix block, so a weak format is handled inside its own budget. Per-question timers remove skipping and returning, which makes a fast partial method valuable: finding one rule and eliminating on it beats finding all three and running out of clock. Adaptive versions pick your next item from your last answer, so early items carry unusual weight and a run of careless quick answers can cap the difficulty, and with it the score, you can reach.

The instructions screen states the scoring rule before you start. Read it: it decides whether your final minute is spent filling blanks or protecting marks.

A worked example: a 3 x 3 matrix

Picture nine cells in a grid, each holding an outlined shape with an arrow inside it and some dots.

Top row, left to right: squares holding three dots, two dots, one dot. Middle row: pentagons holding three, two, one. Bottom row: a hexagon with three dots, a hexagon with two, and the missing cell.

Outlines are unfilled in column one, lightly shaded in column two, solid black in column three. Arrows point up in the top cell of every column and turn a quarter turn clockwise at each step down: right in the middle, down at the bottom row.

Isolate the attributes. Dot count falls by one along each row, so the answer holds one dot. Shape is constant along each row, so it is a hexagon. Shading is fixed by column, so it is solid black. The arrow turns a further quarter turn, so it points down. The answer is a solid black hexagon with one dot and a downward arrow.

Now the distractors. A black hexagon with two dots takes its count from the cell on its left instead of applying the row's decrement to it. A lightly shaded hexagon takes shading from that left neighbour too, where shading is set by column. A black hexagon with an upward arrow turns the quarter turn the wrong way from the cell above. A black pentagon gets everything right but shape, copied from the cell above, where shape is set by row. Each wrong answer satisfies three of the four rules, which is precisely why it looks right.

How to prepare for abstract reasoning specifically

Settle the scoring rule first, because it changes what the last items are worth. If a fraction of a mark comes off for a wrong answer and nothing comes off for a blank, your final minute belongs to items you can narrow to two, not to filling every gap.

Then practise the attribute-isolation pass as a habit rather than a fallback. On every practice matrix, before choosing anything, write out the attributes that vary and the rule governing each.

Score the formats separately. One combined figure across matrices, sequences, odd one out, analogies and set membership hides the thing worth knowing: which mechanism is failing. People strong on sequences and weak on matrices are usually finding one rule and stopping, which is enough on one axis and never enough on two.

Review wrong answers by naming the misapplied rule rather than re-reading the explanation. Ask which rules your option satisfies and which one it breaks. Nearly every error sits in a small set: a row rule read down a column, a second rule missed entirely, a rotation the right way through the wrong angle. Once you can name your own pattern, you can check for it deliberately.

Finally, practise on screen if you will sit on screen. Losing the pencil changes how you hold three rules at once, and people who prepared on paper feel it at once.

Frequently asked questions

TestSolve is independent and not affiliated with any test provider or employer named on this page. All product names and trademarks belong to their respective owners.