Spatial frequency sounds like the kind of term invented to keep outsiders out of the conversation. It just means how fine the stripes are.
If you can keep that translation in your head, the rest of the vocabulary around contrast sensitivity testing falls into place. Why we care about cycles per degree, why a normal human visual system has a peak right around 3, why the curve falls off the same way at both ends — all of it is downstream of being able to picture a striped pattern at three different scales and ask yourself how visible each one is. The rest of this post does exactly that, and ends with the small surprise that the system you look at the world with is unusually well-tuned to faces and edges, and remarkably bad at huge smooth gradients you'd think would be the easiest things to see.
What spatial frequency is
A grating is the simplest possible visual stimulus that is not just a single dot: a smoothly alternating pattern of light and dark bands, like a piece of corduroy held under flat light. A grating has a frequency — how many light-dark cycles fit into some distance.
The thing to watch is the unit on that "some distance." On a printed page, the natural unit is cycles per millimetre. That number is a fact about the paper. But the visual system doesn't see paper. It sees a picture projected onto the retina, and what matters there is not the physical width of the stripes but the angle they subtend at your eye.
So vision science uses cycles per degree of visual angle, abbreviated cpd. One degree of visual angle is about the width of your pinky-finger held at arm's length, or twice the width of the full moon. "1 cpd" means one full light-dark cycle inside that degree; "10 cpd" means ten cycles in the same degree. Cycles-per-degree erases distance dependence — walk towards a striped wall and the stripes fan out (lower cpd); walk away and they crowd into a narrower angle (higher cpd).
At a typical viewing distance of 50 cm, one degree on the screen is about 8.7 mm wide. So 1 cpd means one full light-dark cycle spanning ~8.7 mm; each individual stripe (one bright bar, or one dark bar) is about 4.4 mm. 10 cpd means ten cycles in those same 8.7 mm — each stripe is under half a millimetre. Coarse vs. fine, and now in numbers.
A quick mental analogy. Open any high-resolution photo on your phone, then blur it. As you blur, you have not changed the lighting or colours — you have just deleted the high spatial frequencies and kept the low ones. The face turns into a soft cloud of its overall shape. That progressive deletion of fine pattern is exactly what your visual system does, gently, at the high-frequency end. The blur button is a model.
Seeing the three bands
Here are three patches of the same grating at three different spatial frequencies, drawn to scale for typical screen viewing. The leftmost is low — wide stripes, a few cycles across the patch. The middle is the band your visual system is most sensitive to. The rightmost is fine detail, the edge of resolution for most people on most screens.
Three patches, each at full contrast — same panel size, increasing spatial frequency. Approximate cpd values assume you are viewing the rendered page at about an arm's length on a typical laptop screen; the exact numbers will shift depending on your screen size and how close you are.
Two things to notice as your eye crosses the panels.
First, all three are at the same physical contrast. The black bars are equally dark and the white bars equally bright. There is no luminance trick. What changes between panels is only the spacing.
Second, lean back from the screen. The leftmost panel keeps looking the same — broad enough that distance doesn't matter. The middle panel changes only slightly. The rightmost panel begins to lose its stripes and turn into flat gray as the fine detail outruns your visual system's resolution. Lean forward and the stripes come back. The world hasn't changed; where you sit on the spatial-frequency curve has.
Why your visual system has a sweet spot
Plot a healthy young adult's contrast sensitivity against spatial frequency and you do not get a flat line or a downward slope. You get an inverted U. Sensitivity climbs from moderate at low frequencies to a peak around 3 to 6 cpd, and falls off steeply above that — until it crashes to zero somewhere near 50–60 cpd, the high-frequency cutoff that corresponds roughly to 20/20 acuity. The curve has a name, the contrast sensitivity function (CSF), and it was first laid out formally by Fergus Campbell and John Robson in 1968 (Campbell & Robson, 1968). If you want to see your own CSF drawn directly in an image, see our post on the Campbell-Robson chart; this post is the explainer for why the curve has that shape.
The peak at 3–6 cpd is not arbitrary. The visual system from retina to cortex is built around it.
At the retina, the cells that bundle the photoreceptor inputs (ganglion cells) have centre-surround receptive fields — a small excitatory centre with an inhibitory ring around it. A field shaped like that gives its strongest response to a grating whose bright bar fits in the centre while the dark bars fall on the surround. That tuning makes the retina less interested in very low spatial frequencies (a uniform field excites centre and surround equally and cancels out) and less interested in very high ones (the bars are smaller than the centre and the cell averages over them). The population of ganglion cells across the retina collectively defines a system whose preferred range matches the CSF peak.
In primary visual cortex (V1), individual neurons are tuned to specific spatial frequencies and orientations. De Valois, Albrecht and Thorell mapped that tuning systematically in macaque V1 and found that cells span a wide range of preferred frequencies but the distribution is biased toward the band where psychophysical sensitivity is highest (De Valois, Albrecht & Thorell, 1982). The visual system, in other words, dedicates more neurons to the middle of the spatial-frequency spectrum than to the ends.
What lives in that band? Many of the visual features we care most about: the line of a kerb against pavement at dusk, the edges of letters at reading distance, the contour of a mouth or an arched eyebrow at conversational range. Not the coarse outline of a person across a room (below 1 cpd), and not the eyelash-level detail you only see at close range (above 10 cpd). The system spends its anatomy on what its environment turned out to reward. Owsley's clinical review notes that the shape of the CSF — peak in the middle, falloff at both ends — is robust across healthy populations of all ages; the peak shifts and lowers a bit with age but the inverted U is universal (Owsley, 2003).
Why the high end falls off
Walk through the eye from front to back and you find two honest reasons fine detail eventually disappears.
The first is the optics. The cornea and lens act like a camera lens, and like any lens they have a resolution limit. A faint blur is unavoidable — even an aberration-free eye is limited by diffraction at the pupil, and real eyes have aberrations. The optical "modulation transfer" of a healthy young eye drops noticeably above about 10 cpd and is approaching zero by 50–60 cpd. Add a small uncorrected refractive error or some intraocular scatter (early cataract, dry-eye-related tear-film irregularity) and the optical roll-off moves to lower spatial frequencies — which is why those conditions show up first as high-frequency loss in the CSF.
The second is the photoreceptor mosaic. The light-sensing cones at the centre of your retina are arranged in a roughly hexagonal grid. In the foveola, cone-to-cone centre spacing is about 0.5 arcminutes (about 2.5 µm, where one arcminute is 1/60 of a degree). The information-theory limit on what a regular grid can sample is the Nyquist frequency: roughly one cycle per two grid spacings. Plug in foveal cone spacing and you get a sampling cutoff around 60 cpd. Above that, even if perfect optics delivered a perfect image, the retinal sampling could not distinguish a fine grating from a uniform field. This is why 20/20 acuity (about 30 cpd) sits comfortably inside the retinal sampling limit, and 20/10 — the rare individual with double that resolution — sits roughly at the foveal cone Nyquist. The ceiling is hard, and you can derive it from the geometry of the photoreceptor mosaic.
The story to take away: the high end of the CSF is mostly limited by physics — diffraction, refractive error, scatter, cone spacing — sitting in front of the brain's processing. That is why high-frequency loss is the signature of optical conditions like cataract and uncorrected myopia.
Why the low end falls off
The drop at the low-frequency end is the more surprising one. You might think the visual system would be most sensitive to broad smooth gradients. Why is sensitivity to very low spatial frequencies lower than to mid-frequency stripes?
The answer is lateral inhibition. Retinal ganglion cells respond to local contrast — the difference between centre and surround — not absolute luminance. A uniformly bright field excites the centre and the inhibitory surround equally, and the cell's output is suppressed. The same cell fires vigorously for a sharp edge passing through its receptive field, because the edge breaks the symmetry. This is the architecture behind the Mach bands illusion: at a step from a bright field to a dim one, you see a bright halo on the bright side and a dark halo on the dim side. The retina reports the change, not the steady-state luminance.
That design is why your visual system can work across more than ten orders of magnitude of light level without overflowing. It encodes differences, not absolutes. The cost is that very gradual brightness changes get reported quietly even when their physical contrast is high. A page-sized swatch that smoothly transitions from barely darker than mid-grey to barely lighter will look nearly uniform — even though if you cropped a small square from each end and placed them side by side, the contrast would be obvious.
Low-frequency CSF is not a bug; it is the consequence of a system optimised for detecting change. Reduced low-frequency sensitivity is unusual clinically — the visual system suppresses low frequencies on purpose, but does not completely shut them out, and certain conditions (early glaucoma, some optic-pathway diseases) can degrade the response there beyond the normal architectural reduction.
Why all of this matters for a contrast sensitivity test
A serious contrast sensitivity test probes more than one spatial frequency. The reason has been folded into the explanation above without being said out loud: the conditions that affect vision do not affect the whole curve uniformly. Where the loss shows up on the curve gives you information a single number cannot.
Cataract, with its intraocular scatter, dampens the curve broadly with the heaviest hit at mid-to-high frequencies. Glaucoma, which preferentially damages the magnocellular pathway feeding mid-frequency vision, often produces a sensitivity loss in the 3–6 cpd peak band before visual field tests at the optometrist's office show anything. Multiple sclerosis and related optic-pathway conditions tend to produce broader losses, often heaviest in the mid-to-high range. Concussion and mild traumatic brain injury are associated, in published studies, with disruptions in the mid-frequency band. Uncorrected refractive error shows up almost exclusively at high frequencies. Normal ageing chips away mostly at the high end, declining roughly 10% per decade after age 20.
These are patterns, not diagnoses. None of them tells you which condition you have. But the shape of the curve — where the dip is, how broad it is, whether it sits at the low, peak, or high band — is a real signal that a single Snellen-style acuity measurement cannot give you. That is the whole reason we test multiple frequencies. See contrast sensitivity in MS and vision changes after a concussion; posts on glaucoma and cataract are coming.
Want to see your own curve?
Take the free test. It runs an adaptive procedure across several spatial frequencies, calibrates against your screen and viewing distance, and gives you a number at each frequency. About five minutes. Results stay on your device by default. A reduction at any band is a screening signal worth bringing to a clinician, not a diagnosis on its own.
References
- Campbell, F. W., & Robson, J. G. (1968). Application of Fourier analysis to the visibility of gratings. The Journal of Physiology, 197(3), 551–566. The foundational paper that introduced the contrast sensitivity function and the spatial-frequency-channel framework.
- De Valois, R. L., Albrecht, D. G., & Thorell, L. G. (1982). Spatial frequency selectivity of cells in macaque visual cortex. Vision Research, 22(5), 545–559. Canonical neurophysiology mapping of spatial-frequency tuning in V1; reports that the cortical population is biased toward the band where psychophysical sensitivity peaks.
- Owsley, C. (2003). Contrast sensitivity. Ophthalmology Clinics of North America, 16(2), 171–177. Clinical review of CSF — typical curve shape, age-related changes, and the conditions that produce band-specific losses.
- Watson, A. B., & Pelli, D. G. (1983). QUEST: a Bayesian adaptive psychometric method. Perception & Psychophysics, 33(2), 113–120. Methodology paper underlying the adaptive threshold-estimation procedure most modern contrast-sensitivity tests rely on.