AbilityBenchFree, 4 monocular cues, 60 pairs
Depth perception test built from the cues a flat screen can actually carry
Two shapes appear side by side and you say which is nearer; four different cues take turns being the only difference between them, and each one is scored on its own. It is free, needs no account, no glasses and no calibration step, and it hands back every count it collected. What it cannot do is measure stereo depth, because a display is a single flat surface at a single distance and gives your two eyes nothing to disagree about.
- 100% free
- No signup
- 4 cues x 3 magnitudes
- One eye is enough
- No calibration
Two shapes appear side by side and you say which of them is nearer. Nothing is timed and nothing is hidden after a second: the pair stays on screen until you answer. The first 6 pairs are deliberately ambiguous and are not marked right or wrong — they exist to find out which cue you follow before this page has told you what any of the cues mean.
What a screen can and cannot ask you
Your two eyes sit ≈ 63 mm — Dodgson (2004) apart, and stereoscopic depth comes from the small difference between what each of them sees. A monitor is one flat surface at one distance, so both eyes get the same picture of it and that difference is zero. Every trial here is therefore a pictorial cue — one that works in a painting, a photograph and one open eye. If you want the measurement that needs the two eyes to disagree, it is the stereopsis test, and it cannot start until you have told it your screen size and how far away you are sitting.
Answer with ← and → for the nearer side, or tap the two buttons under the field. One eye is enough for all of it — closing an eye changes nothing here, which is the point.
How to take the four-cue depth test
Answer the contradictory pairs first, learn the conventions second, then read two blocks that answer two different questions.
Answer the first six pairs without knowing the rules
Those pairs carry two cues pointing in opposite directions, and there is deliberately no right answer to any of them. The page withholds what the cues conventionally mean until they are over, because being told that a bigger shape is nearer would replace your reading of the picture with an instruction — and the thing worth recording is which cue you reach for when the picture contradicts itself. Go with whichever impression arrives first rather than reasoning it out.
Read the four conventions, then take the scored block
Between the blocks the page states what each cue means: larger is nearer, the unbroken outline is in front, coarser texture is nearer, the wider sweep is nearer. From then on every pair carries one cue only, at one of three magnitudes, and the smallest of the three is set close to the edge of what anybody can resolve. Answer even when you cannot tell — a forced guess near your limit is the data point the interval is built from, and skipping it just shortens the run.
Read the four rows separately and never add them up
Each cue gets its own proportion correct and its own 95% Wilson interval, plus the split across the three magnitudes. A cue answered well at the coarse step and at chance on the fine one is behaving exactly as a psychophysical function should; a cue at chance on all three was never visible to you at all. A single combined score would hide both patterns, so the page does not compute one.
Technical specifications
| Cues measured | Four, each reduced to one manipulated quantity: relative size as a diameter ratio, occlusion as the fraction of a diameter one disc covers, texture scale as an element-size ratio, and motion parallax as a ratio of horizontal excursion at a shared 1.9 s period |
|---|---|
| Magnitudes per cue | Three. Diameter ratios 1.05, 1.12 and 1.26; overlaps of 3%, 8% and 20% of a diameter; texture scale ratios 1.16, 1.4 and 2.0; excursion ratios 1.16, 1.45 and 2.1 |
| Pairs in a run | Short: 6 contradictory then 24 scored, six per cue. Full: 12 contradictory then 48 scored, twelve per cue and four at each magnitude. Four optional practice pairs sit between the blocks and enter no figure |
| Chance floor | 50% — the answer is one of two sides. The page prints the exact one-sided binomial probability of the whole block against that floor, so 17 of 24 is reported as the one-run-in-31 event it is rather than as 71% |
| Precision beside each score | A 95% Wilson score interval per cue. On the six trials a short run gives one cue that interval spans 39 percentage points at a perfect score and 63 at the middle of the range, so a single cue on a short run settles almost nothing; the full run doubles a cue to twelve trials and still leaves 24 to 49 points, which is why the width is printed beside every score rather than left implied |
| Why no calibration step | Three of the four quantities are dimensionless ratios and survive any screen size, viewing distance or browser zoom. The occlusion overlap is the exception — a fraction of a disc diameter covers a different visual angle on a phone — so the field size and pixel ratio are recorded with that row |
| What is not measured here | Binocular disparity, stereoacuity, vergence and any figure in arcseconds. A shifted image needs a separate view per eye, which an unaided display cannot deliver, so none of those words appears in a result on this page |
| Eyes required | One. Every cue in the battery works in a photograph, so closing an eye changes nothing measurable — which is also the clearest demonstration that a full score here says nothing about whether your eyes work together |
Frequently asked questions
Can a screen test depth perception at all?
It can test the pictorial half of it and not the binocular half, and most pages in this corner of the web do not say which they are doing. Depth information splits into cues that survive a photograph — size, overlap, texture, shading, perspective, parallax — and cues that need two eyes receiving different images. A monitor is a flat plane at one distance, so both of your eyes get the same picture of it and there is nothing left for the second kind. This page builds the first kind properly and sends you elsewhere for the second.
Why does it ask the ambiguous questions before explaining anything?
Because telling you what a cue means destroys the measurement. Once a page has said that the larger of two shapes is nearer, an answer that follows size might be your perception or might be your memory of the sentence, and nothing in the data separates the two. Running the contradictory pairs first, in silence, is the only way to see which cue actually drives your judgment. It is also the reason those six pairs are marked neither right nor wrong: they have no correct answer, and the result reports them as a preference rather than a score.
Occlusion feels trivially easy. Is that a fault in the test?
No, and it is the most informative row on the page. Occlusion is nearly always read correctly and nearly always wins a conflict, because it is the one cue that stays exact at every distance: an object that hides part of another is in front of it, whether the two are ten centimeters or ten kilometers apart. What it will never tell you is by how much, which is why an ordinal cue cannot be turned into a distance. The three magnitudes here shrink the overlap until reading it becomes an acuity problem instead of a depth one, and where your score falls apart marks that transition.
The moving trials — is that really motion parallax?
It is the half of it a browser is allowed to have. Real parallax pairs the sliding retinal image with your own body's record of the movement that caused it, from the vestibular system and from proprioception, and that combination is what makes the depth unambiguous. This page can move the scene but cannot know whether you moved, because it asks the browser for no camera, no microphone and no motion sensor at all. So the trials tell you the convention — the wider sweep is nearer — and then measure how fine an excursion ratio you can resolve, which is a real sensitivity and is not the same thing as unaided parallax depth.
Does a phone give the same result as a monitor?
Three of the four rows transfer directly and the occlusion row does not. Size, texture and excursion are ratios, so shrinking the whole field shrinks both members of every pair and leaves the comparison untouched — that is why the page needs no calibration. The overlap is set as a percentage of the disc diameter, so on a small field it becomes a small number of pixels and a small visual angle, and a 3% overlap that is comfortable on a laptop can be genuinely unreadable on a phone. Compare that row against your own earlier run on the same device.
I have almost no vision in one eye. Is this test meaningless for me?
It is the one depth measurement on this site that still works, which is exactly why it exists. Everything here is monocular by construction: the stimuli were chosen so that closing an eye changes nothing, and people with one working eye read pictorial cues as well as anybody and often better, because they have spent years depending on them. What you should not do is take a strong result here as evidence that stereopsis is intact, or a weak one as evidence that it is not — the two faculties are separate and this page only sees one of them.
Why is there no percentile or age comparison?
Because no published distribution exists for these stimuli, and deriving one from a similar-sounding study would be inventing it. Cue-weighting experiments in the literature use their own displays, their own apertures and their own observers, and a proportion correct on a browser field of unknown size is not comparable to any of them. So the page prints your counts, the chance floor they must beat, and the interval around them, and stops there. A percentile computed from a mean nobody can cite is the one number a visitor would believe and repeat, which makes it the worst thing this site could ship.
What survives on a flat surface, and what leaves with the second eye
Depth information is conventionally sorted into cues, and the sorting matters here because a display keeps some and destroys others. Cutting and Vishton set out the ranking that is still the reference — Cutting & Vishton (1995), Perceiving layout and knowing distances: the interaction, relative potency, and contextual use of different information about depth, in Perception of Space and Motion — and their point is that no cue is best in general, only best at a distance. Occlusion is exact at every range and says nothing about magnitude. Binocular disparity and motion parallax are powerful within a few meters and fade beyond. Relative size and texture gradient carry across the whole range but need an assumption about the world: that the two objects really are the same size, that the surface really is uniformly textured. A screen keeps everything on that list except disparity, and the reason is geometric rather than technical. Your eyes are ≈ 63 mm apart — Dodgson (2004), Variation and extrema of human interpupillary distance, SPIE Stereoscopic Displays and Virtual Reality Systems XI — and stereo depth is built from the difference between the two views that separation produces. A flat panel at a fixed distance presents both eyes with the same surface, so that difference is zero, and no amount of drawing changes it. That is what the stereopsis test has to manufacture with glasses or with free fusion before it can measure anything.
Isolating a single cue turns out to be the hard part of building this, and it is where most browser versions quietly fail. Remove one cue and you often create another: draw two markers at different heights on a receding textured plane and you have added relative height to the visual field; keep their screen sizes equal to control for size and you have told the observer that the far one is physically larger. This page handles it by keeping each pair identical in every respect but one — same vertical position, same outline size, same fill, same stroke — and by naming the residue rather than pretending there is none. The occlusion pair is drawn with matching fills so that the only distinguishing mark is an interrupted arc of stroke, the T-junction that carries the cue; the texture pair keeps both silhouettes the same size so nothing but element scale differs; the moving pair oscillates both discs at one period so the only difference is how far each travels. Landy, Maloney, Johnston and Young called the combination problem weak fusion — Landy et al. (1995), Measurement and modeling of depth cue combination: in defense of weak fusion, Vision Research — and their method is the one the first block here borrows: put two cues in conflict and read the weights off the answers.
The competitive claim of this page is a negative one. Search for a depth perception test and you will find pages that show you a pair of shapes, congratulate you on a percentage and print a figure in arcseconds, and the arcseconds are fabricated: an angle needs a pixel width and a viewing distance, neither of which those pages ever asked for. This one reports ratios, which need no calibration, and refuses the unit entirely. The other measurements on this site that survive an unknown display are the geometric ones — the blind spot test locates your optic disc, the dominant eye test asks which eye you sight with, and the peripheral vision test maps how far out you still catch a change — while the color perception test and the contrast sensitivity test both have to state what an uncalibrated panel costs them. If you would rather see a judgment bent on purpose than measured, the optical illusion test puts a number on how far an illusion moved you. Visitors who arrived here from a hiring pack should know that vision screening for a job is done with plates and a phoropter rather than a browser; the practice pages for the paper-and-pencil part of that pack are the Wonderlic-format practice test and the Mensa-style practice test.
A reaction time here is the interval between the frame that painted the stimulus and the timestamp the browser attached to your key, both read from the same monotonic clock. What neither can see is the display pipeline behind it, so on a 60 Hz screen roughly 16 ms of every figure below is the machine rather than you. That is the timing floor: two numbers closer together than that are the same number, and this page reports no precision it cannot support.
Nothing on this page is timed against a deadline: each pair stays on screen until you answer, so the times reported beside the counts run in seconds and the frame-quantized floor above never binds. It is stated anyway because the same clock is doing the work, and because the one comparison the page draws from those times — contradictory pairs against clean ones — is the distance between two medians, and any distance shorter than the floor is not a distance at all.
This is a measurement exercise, not a clinical assessment. It reports what you did on this page against a stated reference and nothing more — it cannot establish reduced depth perception, a squint, amblyopia or anything else about your eyes. Only a qualified professional, working with more than a browser, can make that judgment.
Where the 60 answers go
Every number on this page is worked out by JavaScript running in the tab you are reading it in. Your answers, your reaction times and your score are never uploaded, logged or kept — which is also why the test carries on working after you disconnect from the network, and why nothing here can be held back behind an email address.
Each answer is a single character in an array in this tab, scored when the run ends and dropped when you close it. Nothing is written to local storage, so a reload starts a fresh run with a fresh shuffle and a finished one cannot be recovered — the copy button exists for that, and it puts the counts, the field size and the run seed on your own clipboard rather than the pair-by-pair record.