How to Solve a 3×3 Matrix Reasoning Question

The shortest reliable solving sequence

A reliable first pass has four moves: describe what changes, propose a rule, predict the missing cell, and check it against another line and every changing feature. When possible, state what the missing tile should contain before looking at the choices. Use choices to test your prediction, not invent a pattern that fits one detail. Consider an original illustration: a three-by-three grid has its lower-right tile missing. Suppose the first row suggests a mark turns one quarter-turn from left to right. Predict how it should appear, then check whether another row supports that relation. Inspect whether a second feature changes independently: matching orientation would not settle a separate count or shading rule. This example is invented and does not reproduce a published test item. This sequence separates observation from explanation. A rule proposes a relation among cells; a prediction states what that relation says the blank should show. *The Raven's Progressive Matrices: A Task Analysis* describes studied items as involving rule induction and applying rules to generate a missing entry. Check whether a proposed relation predicts the blank and survives another comparison. Items differ, so treat rows as evidence, not a universal recipe.

Sources: The Raven's Progressive Matrices: A Task Analysis

What should you inspect before proposing a rule?

Before proposing a rule, make a small inventory of what is visible in each cell. A feature, also called an attribute, is one observable property: for example, the number of marks, their shape, direction, location, fill, or size. A rule is a relation that maps one or more cells to another. Keeping those ideas separate matters because a single glance can bundle several changes together. You may notice that a tile “looks busier” without seeing whether the difference is an extra mark, a darker fill, or both. Naming each property gives you something specific to compare. Use a repeatable scan. First note what stays constant across the cells you can compare. Then list what varies, one feature at a time: count; shape; orientation; position; shading or texture; size; and presence, disappearance, overlap, or combination of elements. “Orientation” means which way a mark faces; “position” means where it sits in the cell. Do not assume every listed property matters. The point is to avoid overlooking one that changes consistently, while leaving irrelevant details out of the eventual rule. A compact note might read: “three marks throughout; one changes direction.” That is observation, not explanation; it keeps a hunch from becoming a conclusion before the comparisons are made. Consider an original illustration, not a test item: imagine a row of three blank cards, each containing small open circles and a short line. From card to card, the line turns a quarter-turn, while the number of circles changes from one to two to three. Record these as two observations: line direction changes; circle count increases by one. Avoid saying “the picture rotates and grows,” which could wrongly suggest that the circles rotate too. Two cards alone cannot tell whether the turn continues or reverses; the third visible card helps distinguish those descriptions. This row suggests a continuing turn and rising count, but says nothing yet about another row or column. For overlapping marks, distinguish the drawn components from the visible regions; a mark hidden behind another may not have been removed. When shapes combine, note which parts remain and which are new. Describe the visible change before assigning a cause: “dark region covers part of the circle” is observation; “the circle is being erased” is an interpretation that needs evidence. The research basis is narrower than a universal puzzle recipe. In “What One Intelligence Test Measures: A Theoretical Account of the Processing in the Raven Progressive Matrices Test,” the authors analyze Raven matrix problems and describe iterative encoding of their regularities. Their task analysis explains that some studied items involve multiple attributes governed by separate relations, and that an answer can satisfy only a subset. That supports checking attributes independently; it does not show that every 3×3 puzzle uses that structure. The review “Cognitive Strategies in Matrix-Reasoning Tasks: State of the Art” surveys varied ways strategy has been studied and says the research advances have been sparse and weakly connected. So treat the inventory as a practical aid, not a scientifically established best sequence. Once your notes distinguish constants from changes, you can test a proposed relation against another comparison instead of trusting the first impression.

Sources: The Raven's Progressive Matrices: A Task Analysis; Cognitive Strategies in Matrix-Reasoning Tasks: State of the Art

Which direction should you test first?

Start with the line that offers the cleanest comparison, not with a rule that you have already decided must govern the whole grid. If the blank is at the lower right, the two completed rows and the two completed columns give you separate places to look for a relation. State what the first line suggests, then ask whether the same relation explains another complete line. The blank’s position changes which comparisons are available. Consider the difference between an edge blank and a central blank before choosing where to begin; the cells surrounding them do not offer the same evidence. At the lower right, every other cell belongs to a complete row or column, so you can infer a pattern from one and check it against the other. At the center, the middle row and middle column each contain the blank, leaving only their outer cells as partial comparisons. You cannot apply a three-cell rule directly to either of those lines because each lacks its third value. A practical selection rule is to compare cells that are easiest to describe precisely. When a line has several competing features, select one at a time and postpone combining them until you know what each comparison supports. Look for a line where the relevant features are clear and stable enough to track: for example, whether a mark changes position, whether a count increases, or whether two elements combine. Then carry that exact statement to a different complete row or column. If it predicts the observed cells there, the second direction has independently supported the idea. You have not proved that no alternative rule could fit, but you have tested the first explanation against more than its starting example. A pattern that fits a single line may be coincidence, or may describe only one of several features. A cross-check tests whether the proposed relationship travels beyond the pair that first suggested it. This cross-check fits research on Raven Progressive Matrices, a particular family of reasoning tests, but that work does not establish a mandatory row-first recipe for every matrix puzzle. In “The Induction of Solution Rules in Raven’s Progressive Matrices Test,” Verguts and De Boeck explain that an item can in principle be solved rowwise, columnwise, or by a combination, while their experiments instructed participants to work rowwise. Their paper also examines how rules used on earlier items can become more available later. The paper concerns Raven tasks under its study conditions, not every puzzle or a universal direction preference. The 1990 analysis “A Theory of Analytic Intelligence” describes an incremental process for Raven problems: compare attributes, infer a possible relation, inspect further entries, and apply the generalized rules to generate the missing entry. This supports a disciplined cycle of observation and checking, rather than a claim that solvers must scan in one fixed order. Start with the clearer line, verify elsewhere, and let a failed check send you back to the relevant cells. That keeps the method efficient while preventing a convenient starting point from becoming an untested assumption.

Sources: The Induction of Solution Rules in Raven's Progressive Matrices Test; The Raven's Progressive Matrices: A Task Analysis

How do you test a rule when several features change?

When a matrix changes in more than one visible way, do not force every change into one rule. Separate the features, infer a relation for each, and combine their predictions only after each relation fits the completed cells. An answer can match the most obvious feature and still fail another part of the pattern. Carpenter, Just, and Shell’s 1990 analysis of Raven Progressive Matrices describes a studied item in which shape, line texture, and orientation follow distinct rules, and notes that some distractors satisfy only a subset. This is evidence about that instrument’s task structure, not proof that every 3×3 puzzle uses the same arrangement. The practical lesson is narrower: check each changing attribute before accepting a tile. Consider this original illustration, unrelated to a published test item. Imagine a grid whose cells contain a small arrow and dots. In each row, the arrow turns one quarter-turn clockwise from left to right. Down each column, the number of dots increases by one. The upper-left cell has an upward arrow and one dot; the next cell in its row has a rightward arrow and one dot; the third has a downward arrow and one dot. In the next row, the first arrow points right and the cells contain two dots; the middle arrow points down with two dots; the last points left with two dots. The bottom row begins with a downward arrow and three dots, then a leftward arrow and three dots. The final cell must contain an upward arrow and three dots. Derive the predictions separately. The horizontal relation determines the arrow: continuing clockwise from down, then left, the third arrow points up. The vertical relation determines the count: each column progresses from one to two to three dots, so the blank in the third row has three. Combining them gives one completion. This is more precise than saying “the shapes rotate and add,” which does not specify which feature changes in which direction or predict the missing cell. Name each feature, state its relation, and record its predicted value. For example: “Arrow orientation advances clockwise across each row; the bottom-right arrow should point up.” Then state: “Dot count increases by one down each column; the bottom-right cell should have three.” Verify both relations against the complete lines. The row supports the orientation rule; the column supports the count rule. The two rules converge without an extra exception. Ask whether changing one predicted feature would break a completed relation. A tile with three dots but a leftward arrow satisfies the count constraint while violating the row’s progression. A tile with an upward arrow and two dots satisfies the orientation constraint but violates the column count. Either is a partial match, not a full solution. Carpenter, Just, and Shell’s 1990 analysis describes partial-rule fits among distractors in the Raven format they studied; that does not mean every test uses identical distractors. If a feature fails its cross-check, pause before adding a special case. Check whether you recorded the feature correctly or reversed the direction, then test another complete row or column. If the relation still fails, discard or revise that hypothesis and seek a simpler account of the observed cells. Verguts and De Boeck’s account of Raven problem solving discusses rowwise, columnwise, and combined approaches, supporting checks in more than one direction for that task context. It does not prescribe a universal order for every matrix. Keep predictions distinct until both fit, then combine them into the missing cell.

Sources: The Raven's Progressive Matrices: A Task Analysis; The Induction of Solution Rules in Raven's Progressive Matrices Test

Should you predict the missing cell before looking at choices?

Usually, make a rough prediction before studying the answer choices, then use the choices to test it. A brief verbal prediction makes your explanation visible and checkable. If you inspect options first, a familiar shape may attract attention before you know which feature matters; the choices can start supplying a pattern you have not inferred from the grid. This is a practical safeguard, not a rule that every item must be solved without looking at choices. State one relation and what it predicts, then compare responses. For multiple properties, predict each separately and check whether candidates meet them all. Matching one conspicuous property is not a complete fit if a second relation fails. Consider an original illustration, unrelated to any published test item. Suppose each cell in a row appears to contain two marks, while their direction changes in sequence. You cannot picture the final direction confidently. The choices show several two-mark arrangements. Eliminate any with one or three marks: it violates the count you inferred, even if its direction looks plausible. Compare the remaining orientations with the row, then check the corresponding column. This is elimination guided by a rule. Choosing an option merely because it resembles the preceding cell is visual matching, not a tested explanation. The methods do different jobs. Prediction asks, “What should be here if my rule is right?” It creates a target that can expose an incomplete idea. Elimination asks, “Which candidates contradict what the grid requires?” It can reject a candidate for violating one feature without requiring you to picture every detail. Neither guarantees correctness: an incomplete prediction can wrongly exclude the answer, while elimination may leave several candidates if the rule is vague. If two options survive, return to the grid and look for a comparison that distinguishes them. Research on Raven Advanced Progressive Matrices supports treating these as distinct parts of solving, but it does not establish a universal tactic for every 3×3 puzzle. In “How knowing the rules affects solving the Raven Advanced Progressive Matrices Test,” researchers supplied some participants with rules for a subset of Raven problems. Across four experiments, supplied rules changed performance and behavior; eye-tracking evidence in the fourth suggested more constructive matching, where solvers build a response from relations instead of only comparing it with choices. Yet working-memory capacity remained related to performance when rules were supplied. The study concerned a particular Raven paradigm: knowing a rule did not remove every demand of managing and applying it. The review “Cognitive strategies in matrix-reasoning tasks: State of the art” describes a varied literature whose findings and methods remain loosely connected. Together, these sources support a limited conclusion: rule knowledge and response strategy matter in studied tasks, but no single order is proven best for all items. Use the sequence conditionally: infer a relation, say what it predicts, inspect options, and test each candidate against the identified features and another available line. If the expected cell is hard to picture, keep the prediction verbal or partial, then let choices help complete it. Revise the rule when an option exposes a contradiction; do not invent an exception to preserve a favored answer. The point is not to avoid choices, but to make them answerable to an explanation grounded in the matrix.

Sources: How knowing the rules affects solving the Raven Advanced Progressive Matrices Test; Cognitive Strategies in Matrix-Reasoning Tasks: State of the Art; The Raven's Progressive Matrices: A Task Analysis

Why a plausible answer can still be incomplete

A rule is stronger when it explains the relevant visible comparisons, predicts the missing cell, and needs few exceptions. If two simple rules still fit, the grid may not distinguish them; keep that uncertainty visible instead of turning a guess into certainty. Find a compact account of the visible marks, then ask whether it makes a testable prediction for this item.

Imagine a grid of abstract tiles in which each tile contains a small symbol and a separate set of short marks. A candidate rule might predict how the symbol changes across a row, while another predicts how the number of short marks changes down a column. These are independent constraints: a proposed completion must satisfy both. Now suppose one answer choice has the expected symbol but the wrong number of marks. It matches the most noticeable feature, yet is incomplete. This illustration is invented to show the reasoning method; it is not a test item or evidence about how often such distractors occur.

A task analysis of Raven Progressive Matrices by Carpenter, Just, and Shell describes items in which attributes such as shape, texture, and orientation can follow distinct rules, and distractors can match only a subset of those rules. Partial matching can mislead: a choice may satisfy one relation while violating another. Its claim is about the analyzed Raven format, not every matrix puzzle. The article’s example above is original and does not reproduce the study’s item.

When two candidate explanations compete, write down what each predicts for the blank before comparing choices. Then check each against every complete row and column that can test it. If one rule fits only by adding a one-off exception, while another explains the same visible features consistently, the second is usually the more useful working hypothesis. This preference for fewer exceptions is a solving heuristic, not a demonstrated psychometric law. Simplicity alone does not prove a rule: a short explanation that fails a completed line must be rejected, even if it feels elegant.

Next ask whether the options separate the predictions. If one candidate predicts a symbol facing one way and another predicts a different orientation, the choices may help decide between them, provided both predictions came from the grid. If every option is compatible with both explanations, or if the blank could be filled in several ways under the visible relations, the item has not supplied a decisive check. Record which feature or comparison remains unresolved; do not silently add a new rule just to select an answer.

The review “Cognitive Strategies in Matrix Reasoning” describes a strategy literature whose developments have been sparse and weakly connected. Its accessible abstract supports caution about treating any one tactic as settled and universal; it does not establish this article’s fewer-exceptions heuristic. In practice, the evidence that should change a tentative answer is concrete: a feature you overlooked, a completed line that contradicts the rule, or an option that reveals your prediction was underspecified. Revise the hypothesis when such a check fails. If no available comparison can decide between two coherent predictions, preserve the ambiguity and move on when the task allows it.

Sources: The Raven's Progressive Matrices: A Task Analysis; Cognitive Strategies in Matrix-Reasoning Tasks: State of the Art

What does practice change when a familiar format gets easier?

Practice can make a familiar matrix format easier by improving how a person notices relations, organizes comparisons, and applies a rule. That is a change in performance on a task with particular demands. A higher score on a repeated or similar matrix test, by itself, does not show that general intelligence increased. “Do We Really Become Smarter When Our Fluid-Intelligence Test Scores Improve?” studied 35 university students who completed short forms of Raven’s Advanced Progressive Matrices (APM) on two days about a week apart. Between sessions, 23 practiced Raven-like problems in two additional sessions; 12 were no-contact controls. Researchers recorded eye movements. More systematic scanning at the second session tended to accompany better scores. Changes in eye-fixation patterns predicted about 30% of the variance in score gains. After statistically accounting for that strategy-related portion, the remaining gain was no longer statistically distinguishable from zero. This makes strategy refinement a plausible contributor to retest gains in this particular task and sample. It does not establish that strategy explains every gain. The design had no transfer group, so it could not directly test whether changes carried over to a different kind of reasoning task. A score can rise because a solver learned to inspect the test more systematically, even when the score is used to discuss a broader ability. A second mechanism is learning which rules are useful and becoming more fluent at applying them. “The Induction of Solution Rules in Raven’s Progressive Matrices Test” examined how experience with earlier items affected choices on later Raven items. In principle, items may be approached rowwise, columnwise, or in combination; their experiments instructed participants to work rowwise. In one experiment, the authors varied how often participants encountered two rule types, then examined whether experience affected which rule they considered later. They report evidence that rules used earlier could become more available and fluent on subsequent items. This supports within-test learning in that context, not a universal rule for solving. Reuse can help when a new item shares a relation with earlier ones: a solver may recognize a familiar change sooner. But availability is not proof of fit. A recent increasing-count item may be followed by one governed by position or orientation. Prior experience suggests a candidate rule; the current grid must confirm it. This distinction clarifies “transfer.” Near transfer is improvement on new items that share important demands with practice, such as inspecting visual relations in another matrix set. Far transfer means improvement on substantially different tasks or a broad ability inferred from them. Better performance on a familiar format is compatible with learning its procedures; it does not establish far transfer. The Hayes, Petrov, and Sederberg study notes that scores also reflect strategy and visuospatial demands, and that test-retest comparisons alone cannot settle whether a broader ability changed. To learn the method, practice with fresh items and review each explanation. Name the relation after a correct response. After an error, locate the failed step: noticing a feature, cross-checking, combining constraints, or checking the prediction. Then try a different item instead of memorizing one answer. Improvement across varied examples can show that the solving procedure is becoming more dependable on those demands. Keeping practice varied matters because repeating one familiar format can reward memory for its surface details; changing examples tests whether the solver can identify and verify relations again. It remains a description of performance, not a conclusion about fixed potential or general intelligence. When a repeated score rises, note whether items were repeated or new, how closely they resembled practice, and whether timing or instructions changed. A claim about broader reasoning would need evidence from other tasks and a design that separates strategy learning from the ability being inferred. Without that evidence, the careful conclusion is that performance improved on the later test, with familiarity or strategy as plausible explanations. That supports continued practice for clearer matrix solving, not a claim of increased general intelligence.

Sources: Do We Really Become Smarter When Our Fluid-Intelligence Test Scores Improve?; The Induction of Solution Rules in Raven's Progressive Matrices Test

When should you stop treating the answer as evidence about yourself?

A correct or incorrect response describes performance on that item under those conditions. It does not establish a person's worth, fixed potential, diagnosis, or normed IQ. Interpret a score only for the use and population supported by evidence for that assessment. Use a matrix answer to learn how you approached that problem; do not turn it into a verdict about who you are. Raw performance means the answers observed on a form. A normed interpretation compares a score with results from a defined reference group, using a documented scoring procedure. Validity concerns evidence for a proposed interpretation and use. These terms answer different questions: what did the person do, how does that result compare with a specified population, and what conclusions are justified? A correct answer alone answers only the first, and only for that item. It does not supply the comparison group or establish that a broader conclusion follows. The Standards for Educational and Psychological Testing frame validity around the intended interpretation and use of scores. Their account recognizes that a score can have multiple meanings, and that examinee population, administration context, rival explanations, and consequences matter. Thus the same observed performance may be useful feedback in practice but inadequate evidence for an education, work, diagnosis, or eligibility decision. The relevant question is whether evidence supports this interpretation for this use and these examinees. The Standards provide a general framework, not an evaluation of every online quiz. They ask test users to consider whether the intended use is supported, rather than assume it from the score's label. This distinction matters after either success or a disappointing miss. One item samples a narrow set of demands: noticing visual features, inferring a relation, tracking a prediction, and responding under the conditions of that session. A miss may reflect an incomplete rule, an overlooked feature, unfamiliarity with the format, or another condition affecting performance. A correct response shows that the answer was reached on that occasion; it does not show how broadly the method generalizes. These are possible explanations to investigate, not diagnoses. Review the reasoning process before drawing a wider conclusion. For Test IQ Free, the boundary is specific. Its adaptive 50-question test provides a provisional estimated IQ range based on a conversion with placeholder parameters and no validated adult reference population. The displayed band is neither a confidence interval nor an established population comparison. That estimate can accompany low-stakes practice feedback, but it cannot carry the meaning of a normed adult IQ score. The optional paid report uses the same provisional conversion and adds explanation; payment does not add validation or accuracy. Neither format establishes giftedness, a clinical condition, or suitability for an education or employment decision. Stop when the conclusion you want to draw exceeds the evidence the test provides. To find which feature you missed, inspect the worked reasoning and try a fresh problem. For a consequential decision about yourself or another person, do not substitute a browser result for qualified assessment and other relevant evidence. A qualified professional can select and interpret an assessment for the actual question and population. A matrix answer is evidence about that response; broader claims need evidence designed to support them.

Sources: Standards for Educational and Psychological Testing

What is a responsible next step after solving?

For low-stakes practice, carry the same sequence to a fresh matrix: state what each changing feature should do, predict the missing cell, then check the prediction against the grid before looking at the choices. Read the worked answer and compare its reasoning with yours. If you missed the item, identify the failed step: perhaps you overlooked a feature, tested only one line, or chose an option before predicting the rule. Practice that operation on a different example instead of memorizing the answer. Test IQ Free’s free adaptive 50-question test is an optional practice route across pattern, number-pattern, and spatial reasoning. Its result includes a provisional estimated IQ range from a conversion with placeholder parameters and no validated adult reference population. This range is not a confidence interval or established comparison with other adults. The optional $9 report adds explanation and a practice plan using the same unvalidated conversion; payment does not make the estimate more accurate or valid. For consequential questions, use a qualified professional assessment suited to the decision and population. The Standards for Educational and Psychological Testing frame score meaning around a proposed interpretation and use, so purpose matters. A browser puzzle can provide practice, but it cannot settle an important decision.

Sources: Standards for Educational and Psychological Testing

Questions readers ask