011 min
Why visual noise hides what people came to see
Picture a team of warehouse coordinators who open the same order table many times a day. They need one fact from it: which orders are running late. The table contains that fact, but it is hard to see.
Every cell has a border. Every second row is shaded grey. Every column label is bold and black, just like the data under it. Five status badges use five bright colours. A banner across the top scrolls news about a new feature.
Nothing on that screen is wrong on its own. Someone added each border, shade and badge for a reason. Together they create visual noise: marks that draw attention but do not help the coordinators with the job in front of them. The late orders are still in the table. The coordinators just take longer to find them, and on a busy day they miss one.
The cost is hard to see in a design review, because no single element causes it. A reviewer who checks the screen one part at a time will approve every part. The problem only shows when someone uses the whole screen for one job, quickly, many times a day.
022 min
How visual noise works
The coordinators are slow because every visible mark asks for a small share of attention. People do not read a screen evenly from corner to corner. Their eyes jump to whatever differs from its surroundings: a strong line, a bright colour, a bold word, something that moves. If only the useful items differ from what is around them, the eye lands on them first. If everything differs, nothing stands out.
Designers borrowed the term from electronics. On a radio or a phone line, the signal is the message and the noise is the random interference mixed into it. The signal-to-noise ratio compares how strong the two are. On a screen, the signal is whatever the person needs for the current task, and the noise is everything else that draws attention. In the order table, the signal is the column that shows how late each order is. The borders, stripes, badges and banner are noise for that job.
Two points follow. First, noise depends on the task. The scrolling banner is signal for someone who wants product news, and noise for a coordinator chasing late orders. Second, noise is not the same as amount. A dense table can stay calm if its extra marks are faint, and a sparse page can be noisy if three items all try to look most important.
Why a new warning cannot stand out on a noisy screen
In 2007, Ruth Rosenholtz, Yuanzhen Li and Lisa Nakano at MIT turned this idea into a number that software can compute from any image. They called their main measure feature congestion. Feature congestion rests on one test: on a more cluttered display, it is harder to add a new item that people will reliably notice. On a calm screen, one red badge is enough. On a screen already full of colours, edges and contrasts, there is no colour or weight left that is different enough.
This explains a failure the coordinators would recognise. Their team adds a red badge for late orders, and it does not help, because the table already uses red, orange and bright blue for other statuses. The new badge is one more colour among many.
Why noisy screens take longer to search
Older lab studies of visual search counted the items on a display: more items, longer search. Real screens have no clean item count, because a border, a shade and a logo are not separate items in the same way. Rosenholtz's measures were built to stand in for that count. The team showed that their clutter scores tracked how long people took to find a target in complex images. In practice, a noisier screen behaves like a screen with more things to check, even when the amount of real content is the same.
031 min
Where the idea of visual noise comes from
The coordinators' table breaks a rule that designers wrote down before most work happened on screens. The rule reached interface design from two directions.
The first was chart design. In his 1983 book The Visual Display of Quantitative Information, the statistician Edward Tufte proposed the data-ink ratio: the ink that shows data, divided by all the ink in the chart. Tufte argued that a chart should remove the ink that does not show data. He called decoration that adds nothing to the data chartjunk: heavy grids, fake 3D effects, pictures behind the bars.
The second was usability. Nielsen Norman Group lists ten rules of thumb, called heuristics, for reviewing an interface. Jakob Nielsen and Rolf Molich wrote the first version of these rules in 1990, and Nielsen revised them in 1994. The eighth heuristic, "Aesthetic and minimalist design", states the mechanism in one sentence:
"Every extra unit of information in an interface competes with the relevant units of information and diminishes their relative visibility."
Visual noise itself is practitioner language, not the name of one research finding. No single person coined it for interfaces. It is the everyday term for the problem Tufte described in charts and Nielsen described in screens, and that perception researchers later measured.
042 min
Visual noise vs. nearby concepts
Several nearby terms describe parts of the coordinators' problem. They are easy to mix up because they often appear together.
| Concept | What it describes | How it differs from visual noise |
|---|---|---|
| Visual clutter | A crowded or disorderly display | Close to a synonym. Clutter stresses amount and disorder. Noise stresses that a mark does not serve the task. |
| Cognitive load | The mental effort a task takes from working memory | Noise is on the screen. Cognitive load is in the person. Noise is one cause of extra load. |
| Information overload | More relevant information than a person can use to decide | Overload is about too much useful content. Noise is about content that is not useful for this task. |
| Visual hierarchy | The order in which a layout tells the eye to look | Hierarchy is the ranking. Noise is what makes the ranking fail. |
| Image noise | Random grain in a photo or camera image | A different meaning of the same words, from photography. |
The last row matters for search. In camera testing and vision science, "visual noise" means random speckle in an image: the grain in a low-light photo, or a noise pattern used in a perception experiment. Camera labs score it by how much grain a person would actually see at a given viewing distance. That meaning is unrelated to interface design.
052 min
How visual noise shows up in real interfaces
The coordinators' table is invented, but the same pattern appears in real products, both fixed and unfixed.
A data grid cleaned up
In a 2009 post, Scott Love of Soliant Consulting cleaned up a small data grid by deleting its vertical lines and its alternating row shading. He argued that the eye can follow the columns of a small grid without vertical rules, and that the shading added clutter rather than help. He then made the grid lines and the column headers grey, so they still guided the eye but no longer competed with the numbers. The data stayed exactly the same. Only the marks around it changed, and the values became the darkest, most visible thing on the screen.
A search page that stays quiet
Google's search home page is a familiar quiet design. It shows a logo, one search box, two buttons and a few small links at the edges. Almost every mark on it serves the one task most visitors arrive with. Compare that with a typical ad-funded article page: a pop-up asking for an email address, a video that starts playing in the corner, side-rail ads, and a row of sharing icons, all around a few paragraphs of text.
Why real content that looks like an ad gets ignored
Readers have learned to defend themselves against this kind of page. The habit is called banner blindness: people skip anything they take to be an ad. In a 2018 eye-tracking study, Kara Pernice and colleagues at Nielsen Norman Group looked at what triggers it. The team found that real content with ad-like traits was skipped too. Placement in a typical ad spot, animation and closeness to real ads were enough. So noise has a second cost. It makes readers stop looking at whole regions of the page, including regions that hold the thing they need.
062 min
How to spot visual noise in a design
A team does not need a lab to find noise like the coordinators'. Most of the signs are visible in normal research and support work:
- In session recordings, people scan the same area several times before they click the main action.
- People miss a status, warning or button that is clearly on the screen.
- Support tickets ask where a setting is, when the setting is visible on the page people were using.
- A blur test fails: blur a screenshot until text cannot be read, and the shapes that still stand out are not the important ones.
- Every new feature request comes with a request to make it stand out more, because nothing on the screen stands out any more.
What clutter ratings actually measure
Survey data adds a warning about what users notice. James Lewis and Jeff Sauro of MeasuringU built a short clutter questionnaire for websites, published in 2024, from 2,761 responses about 57 sites. In their model, overall clutter had about five times as much effect on how good a site looked to people as on how easy they found it to use. People feel noise first as a page that looks bad. It also slows them down, but they report that effect less strongly. So a team that only asks whether a task was easy can miss a clutter problem that users see clearly.
The misreading: noise is only extra content
Many teams treat noise as a problem of too much stuff, and fix it by deleting content. Lewis and Sauro separate two kinds of clutter. Content clutter is irrelevant material, such as ads and autoplay video. Design clutter covers useful content shown badly, such as too little white space, too much text, or a layout that does not follow a clear order. The coordinators' table has almost no irrelevant content. Its noise comes from how the relevant data is drawn, so deleting columns would not fix it.
071 min
When cutting visual noise goes too far
Once the coordinators' designer starts deleting lines and colours, the next risk is deleting too much. Some marks look like decoration but tell people how the screen works. A button border or an underline is that kind of mark. Designers call it a signifier: a visible clue that shows where to act.
What happened when flat design removed the clues
In 2017, Kate Moran at Nielsen Norman Group tested this with eye tracking. Each of 71 participants did a task on one version of nine web pages. One version used strong clues, such as blue underlined links and buttons with clear shapes. The other used weak clues, such as links styled like plain text and buttons drawn only as an outline. People spent 22% more time looking at the pages with weak signifiers. They also fixed their eyes on more points on those pages, a sign that they were searching. The weak clues had made the page look calmer, but they had removed part of the signal along with the noise.
The lesson for the coordinators is direct. The row borders were noise for reading the table. The border around the button that assigns an order to a driver is not. Remove what does not help the task, and keep every mark that tells people what they can do.
082 min
Where the evidence on visual noise is contested
The strict version of the rule, which removes everything that is not data, has been challenged directly. The challenge came from chart research, where the rule began.
When decoration helps memory
In 2010, Scott Bateman and colleagues at the University of Saskatchewan compared plain charts with decorated charts in the style of the graphic artist Nigel Holmes. The decorated charts wrapped the same numbers in large drawings. People described the decorated charts as accurately as the plain ones, and two to three weeks later they remembered the decorated charts significantly better. By Tufte's rule, the drawings were chartjunk. In this study, they did not hurt reading and they helped recall.
Bateman's team did not argue that every chart should be decorated, and the study was small. But it shows where the rule stops. For a chart someone must remember, such as a single figure in a presentation, decoration tied to the topic can be signal. For a table someone reads many times a day, like the coordinators' orders, speed of finding matters more, and the plain version wins.
There is also no single agreed measure of visual noise for interfaces. Rosenholtz's measures work on images, and Lewis and Sauro's questionnaire measures what people report. Until a team has its own data, the safest approach is to decide what the signal is for one named task, and judge every mark against that task.
092 min
Visual noise and accessibility
For some readers, noise is not a matter of speed but of whether they can use the page at all. The Web Content Accessibility Guidelines (WCAG) 2.2 set rules on both sides of the problem.
Moving content
WCAG Success Criterion 2.2.2 Pause, Stop, Hide covers anything that moves, blinks or scrolls for more than five seconds, starts by itself and sits next to other content. People must be able to pause, stop or hide it. The W3C notes that blinking content makes it hard for people with attention deficit disorders to focus on the rest of the page. The coordinators' scrolling banner fails this rule unless it has a pause control.
Greyed-out text that fails contrast
Turning labels grey is a common way to quiet them, and it can go too far. Under Success Criterion 1.4.3 Contrast (Minimum), the rule is simple. Normal-size text needs at least 4.5:1 contrast with its background. Large text needs at least 3:1. A light grey that makes a label pleasantly quiet on a good monitor can make it unreadable for someone with low vision.
Inputs that lose their borders
Removing borders is the other common move, and form fields are where it breaks. Success Criterion 1.4.11 Non-text Contrast covers this case. Any visual part that people need in order to see that a control is there, such as the border of a field, needs 3:1 contrast with the colours next to it. A field with no visible edge at all may not be recognised as a field.
102 min
How to reduce visual noise
With the limits in place, the coordinators' designer can now work through the table. The order of the steps matters, because each one makes the next easier to judge.
Name the task and its signal.
Write down the one job the screen must support first, and the data that job needs. For the coordinators, that job is finding late orders, and the signal is the lateness column.
Remove marks that do the same job twice.
Borders and row shading both separate rows. White space can do it alone, because people see items that sit close together as one group. The designer removes the cell borders, drops the grey row shading, and turns the column labels grey.
Mute what must stay.
Keep labels, rules and secondary data, but make them lighter or smaller than the signal, within WCAG contrast limits.
Spend strong colour once.
Use one accent colour for the one state that needs action. The late rows get it. Other statuses become text or neutral icons. Keep to one or two type styles, so that bold text still means something.
Stop motion that is not about the task.
Pause the banner, or move release news to a place people visit when they want it.
Check the clues.
Before shipping, confirm that every control still looks like a control.
Check the phone layout.
On a small screen each extra mark takes a larger share of the space, so noise that is tolerable on a desktop becomes a problem on a phone.
None of these steps is large on its own. After them, the table holds the same data. The difference is that the late orders are now the first thing anyone sees. Use progressive disclosure for detail that only some people need, rather than deleting it.
Do not apply this to every screen in the same way. A screen people visit once, such as a launch page, may need images and character that a daily work tool does not. Judge each mark against the task on that screen, and keep the marks that serve it.
?4 questions
Questions people ask
Is white space a kind of visual noise?
Can a dense screen have little visual noise?
Who decides what counts as noise on a screen with many users?
Does dark mode add or remove visual noise?
§13 sources
Sources
Show all 13 sourcesHide sources
Ruth Rosenholtz, Yuanzhen Li and Lisa Nakano, "Measuring visual clutter", Journal of Vision 7(2), 2007 (PubMed record).
Jakob Nielsen, "10 Usability Heuristics for User Interface Design", Nielsen Norman Group (first version 1990, revised 1994).
Scott Bateman, Regan L. Mandryk, Carl Gutwin, Aaron Genest, David McDine and Christopher Brooks, "Useful Junk? The Effects of Visual Embellishment on Comprehension and Memorability of Charts", CHI 2010.
Patrick J. Lynch and Sarah Horton, "Signal-to-noise ratio", Web Style Guide, 3rd edition (on Tufte's data ink and chartjunk).
Scott Love, "Visual Noise", Soliant Consulting, 2009.
Kate Moran, "Flat UI Elements Attract Less Attention and Cause Uncertainty", Nielsen Norman Group, 2017.
Kara Pernice, "Banner Blindness Revisited: Users Dodge Ads on Mobile and Desktop", Nielsen Norman Group, 2018.
James R. Lewis and Jeff Sauro, "Measuring the Perceived Clutter of Websites", International Journal of Human-Computer Interaction, 2024 (author copy).
James R. Lewis and Jeff Sauro, "Incorporating Clutter in the SUPR-Q", MeasuringU.
James R. Lewis and Jeff Sauro, "Exploratory Development of a Clutter Questionnaire for Websites", MeasuringU.
W3C, "Understanding SC 2.2.2 Pause, Stop, Hide", WCAG 2.2.
W3C, "Understanding SC 1.4.3 Contrast (Minimum)", WCAG 2.2.
W3C, "Understanding SC 1.4.11 Non-text Contrast", WCAG 2.2.





