Attention
Thatcher Illusion Task
Also called: Thatcher effect, Thatcher illusion test, thatcherized faces task, thatcherised faces, Margaret Thatcher illusion
Measure how turning a face upside down hides a change: participants see two pictures of a face in a row and say whether they were the same, scored by how much worse changes to the eyes and mouth are spotted upside down.
6-8 minUntimed
Free to start: 10 free participants per account, then $0.99 per participant for academic research, or any yearly plan. Try it here first, with no account.
- Papers, 2021 to 2025
- 18
- Evidence grade
- D
- Registered clinical studies
- 0
What it measures
Measures how much turning a face upside down hides a change to its eyes and mouth. Participants see two pictures of one face in a row and say whether they were the same, upright and upside down. The main score is the inversion cost: how much worse changes are spotted upside down.
Constructs
- Configural face perception
- Face inversion effects
- Sensitivity to the orientation of facial features
Research uses
- Orientation-specific face processing in research on development, aging and expertise
- Face perception research in autism and prosopagnosia, at the group level
- Batteries of face tasks with the part-whole and composite face effects
How the paradigm works
The Thatcher illusion task measures how much turning a face upside down hides a change to its eyes and mouth. Thompson (1980) turned the eyes and mouth of a photograph upside down within the face: upright the change is striking, but with the whole picture upside down it is hard to see. Each round shows two pictures of one face, one after the other, for 800 ms each and 200 ms apart, and the participant answers whether they were the same, with no time limit (Psalta, Young, Thompson and Andrews, 2014). In half the rounds the two pictures are the same; in the other half one of them has the eyes and mouth turned 180 degrees. Half the rounds are upright and half upside down, mixed in one session. Psalta and colleagues found that people noticed the change in only 15.6% of upside-down pairs, while upright they did as well as on pairs of identical pictures.
More about the paradigm
The photographs come from the Face Research Lab London Set (DeBruine and Jones, 2017): 12 people, 6 women and 6 men, in grayscale in an oval that hides hair, ears and clothing, each shown as photographed and with the eyes and mouth turned by our code. Every person appears in each kind of pair upright and upside down: 96 rounds. Four practice rounds with feedback come first.
The main score is the inversion cost: d prime for telling changed from unchanged pairs upright minus upside down, a hit being Different on a changed pair and a false alarm Different on a same pair, with the log-linear correction (Hautus, 1995). The task also reports d prime in each orientation, the percent of changes spotted, the percent of same pairs answered Different, accuracy and the median answer time. Answers within 100 ms of the buttons are counted as anticipations and left out. Scores support research and do not diagnose.
Scores
- Inversion Costd'
- d prime upright minus d prime upside down, for telling changed from unchanged pairs. Positive means the change was harder to see upside down.
- Informational; an effect of face orientation
- Upright Sensitivityd'
- z(hit rate) minus z(false-alarm rate) on upright pairs: a hit is Different on a changed pair, a false alarm Different on a same pair, both rates corrected (Hautus, 1995).
- Higher means changes and same pairs were told apart better
- Upside-Down Sensitivityd'
- The same on upside-down pairs.
- Higher means changes and same pairs were told apart better
- Upright Changes Spotted%
- Percent of upright changed pairs answered Different.
- Informational
- Upside-Down Changes Spotted%
- Percent of upside-down changed pairs answered Different.
- Informational
- Upright False Alarms%
- Percent of upright same pairs answered Different.
- Quality check
- Upside-Down False Alarms%
- Percent of upside-down same pairs answered Different.
- Quality check
- Accuracy%
- Percent of answered rounds answered right, both orientations.
- Quality check
- Median Answer Timems
- Median time from the buttons appearing to the answer, over answers at or after 100 ms.
- Informational; the answer has no time limit
- Rounds Answeredtrials
- Rounds answered in the scored run, leaving out anticipations.
- Fewer than planned is flagged for review
- Planned Roundstrials
- Rounds the session planned: 96.
- Informational
- Anticipationscount
- Answers within 100 ms of the buttons appearing, left out of the other scores.
- Quality check
Use and evidence
18 papers from 2021 to 2025 name this task in their title or abstract, up 29% compared with 2016 to 2020. That ranks it 410 of 495 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 2 |
| 2024 | 3 |
| 2023 | 9 |
| 2022 | 1 |
| 2021 | 3 |
| 2020 | 1 |
| 2019 | 2 |
| 2018 | 1 |
| 2017 | 4 |
| 2016 | 6 |
| 2015 | 3 |
| 2014 | 12 |
| 2013 | 10 |
| 2012 | 6 |
| 2011 | 6 |
| 2010 | 15 |
| 2009 | 8 |
| 2008 | 5 |
| 2007 | 5 |
| 2006 | 3 |
| 2005 | 3 |
| 2004 | 3 |
| 2003 | 5 |
| 2002 | 3 |
| 2001 | 3 |
| 2000 | 1 |
Use grade E
Fewer than 30 papers from 2021 to 2025.
Evidence grade D
Little published measurement evidence found.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 2
- Registered clinical studies using it as an outcome
- 0
How the grades work
Use: A is 1,000 or more papers from 2021 to 2025, B 300 to 999, C 100 to 299, D 30 to 99 and E fewer than 30. Evidence: A means at least 25 papers report its test-retest reliability and at least 25 registered clinical studies use it as an outcome; B, at least 8 test-retest papers, and either 5 registered clinical studies or 150 reliability or validity papers; C, at least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies; D, little published measurement evidence found. The counts measure use. They include every paper that names the task in its title or abstract. Grades describe the published research on this task, not ConductCognition's version of it.
Source: ConductCognition test census 0.22.0, counted 2 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Faces
- Grayscale photographs of 12 adults from the Face Research Lab London Set, 6 women and 6 men, aligned on the eyes and cropped to an oval that hides hair, ears and clothing, up to 280 by 320 px, on a black panel. Each person also appears with the eyes and the mouth turned 180 degrees within the face.
- Each round
- A 500 ms cross, the first picture for 800 ms, a 200 ms blank and the second picture for 800 ms, then Same or Different, by click or tap, with no time limit. Answers within 100 ms of the buttons are counted as anticipations and left out. A 500 ms blank follows each answer.
- Pairs
- Both pictures show the same person: the photograph twice, the changed picture twice, or one of each in either order, both upright or both upside down.
- Rounds
- 96: each person in each kind of pair upright and upside down, half the same and half different, in an order drawn for each session with no person twice in a row and no more than 4 of one answer in a row. A rest after 48.
- Practice
- 4 rounds of 2 people the test does not use, upright and upside down, with feedback for 1,500 ms.
The procedure follows the same-or-different task of Psalta, Young, Thompson and Andrews (2014), 800 ms for each picture and 200 ms between them, with photographs of people the participant does not know where they used famous faces, and both orientations mixed in one session where they ran them apart. The score is the inversion cost in d prime; Psalta and colleagues reported accuracy. No screen names or labels a face, and the question is only whether two pictures were the same. Its people include 10 of the Face Matching Task's 40, 4 of the Face Memory Test's 24 distractor faces, 6 of the Face-name Memory Task's 24 foils and 2 of the Chimeric Faces Task's 18, so in a study with the Face Matching, Face Memory or Face-name Memory test, run this task after them or in a separate session. The Part-Whole Face Task uses 4 of its people for the outline of its composite faces, the same in both choices, so the two tasks can run in either order.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: thatcher_illusion_test from the scored run, one Same or Different answer per pair of pictures with no time limit (chance 1 in 2). Accuracy significantly below chance invalidates and not above chance is flagged. One answer on 90% of pairs invalidates when accuracy is not above chance; half the pairs are the same and half differ. Answers under 200 ms on at least half the pairs invalidate. A run short of its planned rounds is flagged for review, and so is spotting the change in fewer than 50% of the upright changed pairs.
Why these rules
Accuracy significantly below chance (one-sided exact test, p < .05) cannot come from guessing; it means the response mapping was reversed or the task was not followed (Slick, Sherman and Iverson 1999 treat below-chance forced-choice performance as definite response bias). Chance-level accuracy can be a real result in impaired participants, so it is only flagged. One option on at least 90% of answered trials, when the correct options are balanced and accuracy is not above chance, means the answers did not depend on the items (the task counterpart of long-string responding in surveys; Curran 2016). An answer under 200 ms is faster than a simple visual reaction, so it cannot reflect the item. Following response-time effort screening for untimed tests (Wise and Kong 2005), the session is invalid only when at least half the answers are that fast. Upright, the change to the eyes and mouth is easy to see (Psalta and colleagues, 2014), so missing most upright changes means the pictures were not watched, and the two pictures show for set times, so a hidden tab withholds the percentile.
No percentile. Scores and quality flags only.
Running this test
Settings you can change
This test runs with fixed settings.
Need a different version of this test?
Devices and languages
- Untimed test: runs on desktops, laptops, tablets and phones. Participant devices
- Languages: English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese.
- Available on: Every plan, including the free plan.
- No percentile. Scores and quality flags only. How norms work
- Recruiting online? See the setup guides for Prolific, Sona Systems, CloudResearch Connect and Qualtrics.
Cite and describe
Participants completed the Thatcher Illusion Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Thompson, 1980). The main score was inversion cost (d prime upright minus d prime upside down, for telling changed from unchanged pairs).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Thompson, P. (1980). Margaret Thatcher: A new illusion. Perception, 9(4), 483-484. https://doi.org/10.1068/p090483
- Psalta, L., Young, A. W., Thompson, P., & Andrews, T. J. (2014). The Thatcher illusion reveals orientation dependence in brain regions involved in processing facial expressions. Psychological Science, 25(1), 128-136. https://doi.org/10.1177/0956797613501521The same-or-different task: 800 ms for each picture, 200 ms apart; the change noticed in 15.6% of upside-down pairs.
- Donnelly, N., Zürcher, N. R., Cornes, K., Snyder, J., Naik, P., Hadwin, J., & Hadjikhani, N. (2011). Discriminating grotesque from typical faces: Evidence from the Thatcher illusion. PLoS ONE, 6(8), e23340. https://doi.org/10.1371/journal.pone.0023340Telling a changed from an unchanged picture is slower and less accurate upside down, and rests more on the eyes than the mouth.
- Mestry, N., Menneer, T., Wenger, M. J., & Donnelly, N. (2012). Identifying sources of configurality in three face processing tasks. Frontiers in Psychology, 3, 456. https://doi.org/10.3389/fpsyg.2012.00456Judging whether the eyes and mouth were turned in 120 ms displays: d prime 1.89 upright against 0.52 upside down.
- Utz, S., & Carbon, C.-C. (2016). Is the Thatcher illusion modulated by face familiarity? Evidence from an eye tracking study. PLOS ONE, 11(10), e0163933. https://doi.org/10.1371/journal.pone.0163933One face at a time: 91.4% right upright against 72.3% upside down.
- Hautus, M. J. (1995). Corrections for extreme proportions and their biasing effects on estimated values of d'. Behavior Research Methods, Instruments, & Computers, 27(1), 46-51. https://doi.org/10.3758/BF03203619The log-linear correction behind d prime.
- DeBruine, L., & Jones, B. (2017). Face Research Lab London Set [Data set]. figshare. https://doi.org/10.6084/m9.figshare.5047666.v5
Source papers
- Thompson P (1980). Margaret Thatcher: A new illusion. Perception, 9(4):483-484.Paradigm source, Adults
- DeBruine L, Jones B (2017). Face Research Lab London Set. figshare. Dataset. doi:10.6084/m9.figshare.5047666.v5Paradigm source, Adults
Questions
How long does the Thatcher Illusion Task take?
About 6 to 8 minutes, as listed in the test library.
What device does a participant need?
Untimed test: runs on desktops, laptops, tablets and phones. The browser must be Chrome, Edge, Firefox or Safari.
What scores does it produce?
Inversion Cost, Upright Sensitivity, Upside-Down Sensitivity, Upright Changes Spotted, Upside-Down Changes Spotted, Upright False Alarms, Upside-Down False Alarms, Accuracy, Median Answer Time, Rounds Answered, Planned Rounds and Anticipations. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Thatcher Illusion Task used in research?
18 papers from 2021 to 2025 name it in their title or abstract, up 29% compared with 2016 to 2020. That ranks it 410 of 495 test families in the ConductCognition test census (use grade E). Source: ConductCognition test census 0.22.0, counted 2 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Thatcher Illusion Task have?
In the published research, 0 papers report its test-retest reliability and 0 registered clinical studies use it as an outcome (evidence grade D in the ConductCognition test census).
Does it report a percentile?
No. Scores and quality flags only.
Which languages can participants use?
Participants can take this test in English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese. English is on every plan, and the other languages come with yearly plans. Have a native speaker read the machine-translated text before launch.
Is this a clinical test?
No. ConductCognition is for research use only. It runs, scores and exports the test; clinical interpretation is outside the platform.
Every test: the full test list.
Run this test in your next study
Start free with 10 free participants on any test, then use participant credits or a yearly plan.