psychology.games

Hall II · Psychological Phenomena, Tasks & Thought Experiments

Wason selection task

Station · Playable

A puzzle with a normatively correct answer — and being wrong is the point of the page. Nothing is scored or stored.

Puzzle 1 of 2

Rule to check: “If a card has a vowel on one side, it has an even number on the other side.”

Which cards must you flip to find out whether the rule is being broken?

The experience

Four cards lie in front of you: E, K, 4, 7. Each has a letter on one side and a number on the other, and you are handed a rule to test — every vowel card has an even number on its back. Your job is to turn over only the cards that could tell you whether the rule holds. Before you choose, it helps to know how this usually goes: shown four cards and a rule like "if a card has a vowel on one side, it has an even number on the other," most people check the vowel card and the even-number card — and skip the odd-number card, the one that could actually break the rule. In Wason's original studies almost everyone picked the named cards, and across hundreds of later experiments only somewhere between about 4% and 20% of people choose the two logically correct cards on this abstract version.Convergent / replicated If you felt the pull toward E and 4, you were in very large company.

What is happening

The strange part is not that the puzzle is hard — it is that a cosmetic rewrite makes it easy. Dress the same card structure up as a rule to enforce — "if a person is drinking beer, then the person must be over 19," with cards for a beer drinker, a coke drinker, a 16-year-old, and a 22-year-old — and suddenly most people get it right, checking the beer drinker and the 16-year-old. Across such deontic (rule-enforcement) versions, about two-thirds of participants make the correct selection, versus roughly one in five on abstract versions.Convergent / replicated That is why the station above deals you both puzzles: first the letters and numbers, then the bar (our version simply sets the drinking age at 18). The logical skeleton is identical in the two rounds; only the clothing changes, and with it, what your mind spontaneously checks.

What this shows

Why should a story about a bar unlock a problem that letters and numbers lock up? One influential answer looks to our social history rather than our logic training. Leda Cosmides proposed that people do well on rule-enforcement versions because human minds carry evolved machinery for spotting cheaters in social exchanges — deals of the form "if you take the benefit, you must pay the cost." When the rule is a social contract and the task is to find violators, 65–80% of people answer correctly, even with completely unfamiliar rules, and the pattern has been reported across very different societies, from US students to Shiwiar hunter-horticulturalists in Amazonian Ecuador.Supported, with limits Whatever the final explanation turns out to be, the two rounds you just played make one thing vivid: what you think a rule is for shapes which evidence you go looking for.

What this does not show

Turning over E and 4 puts you with the majority, not with the careless. Missing the odd-number card does not show that a person is broadly illogical or unintelligent: the very same rule structure is solved by most people when it is framed as a rule to enforce, so the failure is tied to how the problem is interpreted, not to a fixed inability to reason. Oaksford and Chater's Bayesian analysis goes further, arguing the typical "wrong" picks can be sensible evidence-gathering if people treat the task as asking which cards are most informative about an uncertain rule — so the task should not be read as an IQ-style test of general rationality.Supported, with limits Nothing on this page is scored or stored, and getting the abstract round "wrong" tells you about a famous pattern in human reasoning — not about your intelligence.

Research and limitations

Peter Wason introduced the four-card task in the 1960s, and it has since become one of the most intensively studied problems in the psychology of reasoning. What it has not become is settled. Why wording and content change performance so dramatically is still argued over: rival explanations include learned pragmatic "permission" schemas (Cheng & Holyoak), evolved cheater detection (Cosmides), relevance-driven attention, and rational information-seeking, and one meta-analysis counts at least 15 distinct theories of the task. Sperber and colleagues showed they could make people succeed on non-social, descriptive versions just by changing the context that makes the falsifying cases relevant — a direct challenge to explanations tied specifically to social or deontic rules.Mixed / contested

So treat the two puzzles above as a demonstration with a firm surface and a contested core: what people reliably select is well documented across hundreds of experiments, while the story about why is still being argued over. Where the research is silent — for instance, on what any single person's picks say about them — this page stays silent too.

Research drawer — claims, sources and limits

Every highlighted claim above is recorded here with its evidence grade and sources. Grades are per claim; this site never computes an overall truth score.

↩ shown four cards and a rule like "if a card has a vowel on one side, it has an even number on the other," most people check the vowel card and the even-number card — and skip the odd-number card, the one that could actually break the rule. In Wason's original studies almost everyone picked the named cards, and across hundreds of later experiments only somewhere between about 4% and 20% of people choose the two logically correct cards on this abstract version.

Convergent / replicated

Wason, P. C. (1968). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20(3), 273–281. https://doi.org/10.1080/14640746808400161 · accessed 2026-08-16
Support note: Read in full (open PDF mirror at web.mit.edu). Introduces the four-card task and coins the name "selection task." Describes the vowel/even-number pilot version and uses the rule "If there is a D on one side of any card, then there is a 3 on its other side." States that in the pilot and Hughes's replication "Nearly all subjects select P, from 60 to 75 per cent select Q, only a minority select Q-bar [not-Q] and hardly any select P-bar," calls the two errors "enduring and deep," and reports two "therapy" experiments (N=36 UCL students) in which prompting subjects to project falsity failed to reli

Ragni, M., Kola, I., & Johnson-Laird, P. N. (2017). The Wason Selection Task: A Meta-Analysis. Proceedings of the 39th Annual Meeting of the Cognitive Science Society, 980–985. https://escholarship.org/uc/item/3rv0k45d · accessed 2026-08-16
Support note: Read in full (open PDF). Meta-analysis of 228 selection-task experiments (104 abstract, 44 everyday, 80 deontic; over 300 published experiments in 50 years are noted). Table 2 gives the percentages of the four canonical selections for abstract rules: p alone 36%, p&q 39%, p,q&not-q 5%, and the correct p&not-q only 19%. Also documents that participants "tend to overlook one of the cards that could falsify the rule."

↩ Dress the same card structure up as a rule to enforce — "if a person is drinking beer, then the person must be over 19," with cards for a beer drinker, a coke drinker, a 16-year-old, and a 22-year-old — and suddenly most people get it right, checking the beer drinker and the 16-year-old. Across such deontic (rule-enforcement) versions, about two-thirds of participants make the correct selection, versus roughly one in five on abstract versions.

Convergent / replicated

Griggs, R. A., & Cox, J. R. (1982). The elusive thematic-materials effect in Wason's selection task. British Journal of Psychology, 73(3), 407–420. https://doi.org/10.1111/j.2044-8295.1982.tb01823.x · citation only
Support note: Best-known citation for the drinking-age problem, which produced high correct-selection rates where matched abstract versions produced almost none. Not opened directly (Wiley returned HTTP 402 paywall); however, the opened Sperber, Cara & Girotto (1995) paper reproduces the exact Griggs & Cox drinking-age problem as its Fig. 2 (police-officer framing, rule "if a person is drinking beer, then the person must be over 19 years of age") and states that on deontic versions "a majority of subjects correctly selects the P and the not-Q card."

Ragni, M., Kola, I., & Johnson-Laird, P. N. (2017). The Wason Selection Task: A Meta-Analysis. Proceedings of the 39th Annual Meeting of the Cognitive Science Society, 980–985. https://escholarship.org/uc/item/3rv0k45d · accessed 2026-08-16
Support note: Read in full. Quotes the drinking-age rule (citing Griggs & Cox, 1982) as the exemplar deontic version. Table 2: across 80 deontic experiments, 64% of participants made the correct p&not-q selection, versus 19% across 104 abstract experiments; the deontic task also showed the highest cross-experiment concordance (Kendall's W = .54).

↩ Leda Cosmides proposed that people do well on rule-enforcement versions because human minds carry evolved machinery for spotting cheaters in social exchanges — deals of the form "if you take the benefit, you must pay the cost." When the rule is a social contract and the task is to find violators, 65–80% of people answer correctly, even with completely unfamiliar rules, and the pattern has been reported across very different societies, from US students to Shiwiar hunter-horticulturalists in Amazonian Ecuador.

Supported, with limits

Cosmides, L. (1989). The logic of social exchange: Has natural selection shaped how humans reason? Studies with the Wason selection task. Cognition, 31(3), 187–276. https://doi.org/10.1016/0010-0277(89)90023-1 · citation only
Support note: Original statement of social contract theory tested with selection tasks, including unfamiliar social-exchange rules; per its structure (confirmed via Semantic Scholar metadata and search summaries; PMID 2743748), Part I tests social exchange theory against availability theories and Part II against Cheng & Holyoak's permission-schema theory. Not opened: publisher page blocked and PubMed abstract page was cookie-walled.

Cosmides, L., Barrett, H. C., & Tooby, J. (2010). Adaptive specializations, social exchange, and the evolution of human intelligence. Proceedings of the National Academy of Sciences, 107(Suppl. 2), 9007–9014. https://doi.org/10.1073/pnas.0914623107 · accessed 2026-08-16
Support note: Read in full (open PDF at cep.ucsb.edu). States that with indicative (descriptive) conditionals "only 5–30% of normal subjects respond with the logically correct answer, P & not-Q," whereas "when the conditional rule involves social exchange and detecting a violation corresponds to looking for cheaters, 65–80% of subjects correctly detect violations," with no difference between familiar and totally unfamiliar social contracts; reports the pattern "in every culture where it has been tested," including Shiwiar hunter-horticulturalists, and reports experiments arguing violation-detection tracks i

↩ Why wording and content change performance so dramatically is still argued over: rival explanations include learned pragmatic "permission" schemas (Cheng & Holyoak), evolved cheater detection (Cosmides), relevance-driven attention, and rational information-seeking, and one meta-analysis counts at least 15 distinct theories of the task. Sperber and colleagues showed they could make people succeed on non-social, descriptive versions just by changing the context that makes the falsifying cases relevant — a direct challenge to explanations tied specifically to social or deontic rules.

Mixed / contested

Sperber, D., Cara, F., & Girotto, V. (1995). Relevance theory explains the selection task. Cognition, 57(1), 31–95. https://doi.org/10.1016/0010-0277(95)00666-M · accessed 2026-08-16
Support note: Read (open PDF at dan.sperber.fr, pp. 31–34 examined in detail). Abstract states the relevance-based account: subjects infer testable consequences in order of accessibility and stop when the interpretation meets their expectations of relevance, and "by devising appropriate rule-context pairs, we predict that correct performance can be elicited in any conceptual domain," corroborated in four experiments. The introduction documents the rival theory landscape (citing Cheng & Holyoak 1985; Cosmides 1989; Oaksford & Chater 1994, among others), notes the task is "the most intensively researched sing

Ragni, M., Kola, I., & Johnson-Laird, P. N. (2017). The Wason Selection Task: A Meta-Analysis. Proceedings of the 39th Annual Meeting of the Cognitive Science Society, 980–985. https://escholarship.org/uc/item/3rv0k45d · accessed 2026-08-16
Support note: Read in full. States there are "at least 15 distinct theories of the selection task," including ones based on the meaning of conditionals, formal rules, matching heuristics, content-specific rules of inference, and probabilities (Oaksford & Chater, 1994), and calls the persistence of 15 theories after half a century "embarrassing for cognitive science." Its entropy analysis rules out the whole class of theories that assume cards are selected independently of one another, leaving only two theories that predict selection frequencies.

↩ Missing the odd-number card does not show that a person is broadly illogical or unintelligent: the very same rule structure is solved by most people when it is framed as a rule to enforce, so the failure is tied to how the problem is interpreted, not to a fixed inability to reason. Oaksford and Chater's Bayesian analysis goes further, arguing the typical "wrong" picks can be sensible evidence-gathering if people treat the task as asking which cards are most informative about an uncertain rule — so the task should not be read as an IQ-style test of general rationality.

Supported, with limits

Oaksford, M., & Chater, N. (1994). A rational analysis of the selection task as optimal data selection. Psychological Review, 101(4), 608–631. https://doi.org/10.1037/0033-295X.101.4.608 · accessed 2026-08-16
Support note: Read (open PDF at pages.ucsd.edu, pp. 608–609 examined in detail). Abstract: performance "has been assessed against a now outmoded falsificationist philosophy of science"; their Bayesian model of optimal data selection "fits well with people's performance on both abstract and thematic versions of the task" and "suggests that reasoning in these tasks may be rational rather than subject to systematic bias." Also documents the standard abstract-task data (from Johnson-Laird & Wason, 1970a): only 4% select p and not-q, versus p&q 46% and p alone 33%, and introduces the rarity assumption (propertie

Ragni, M., Kola, I., & Johnson-Laird, P. N. (2017). The Wason Selection Task: A Meta-Analysis. Proceedings of the 39th Annual Meeting of the Cognitive Science Society, 980–985. https://escholarship.org/uc/item/3rv0k45d · accessed 2026-08-16
Support note: Read in full. Its Tables 1–2 quantify the content-dependence relied on by this claim: the identical selection structure yields 19% correct (p&not-q) with abstract rules but 64% with deontic rules across 228 experiments, demonstrating that failure on the abstract version does not generalize to rule-checking with realistic enforcement content.