Optimizing UI icons for faster recognition

Posted by

What makes an icon a valuable addition to the interface, rather than a mere decorative element? Intuitiveness, aesthetic value, memorability, intercultural perception? While an effective icon would combine many of those characteristics, I’d like to focus on one measure–speed of recognition, or how fast a specific icon can be discovered and identified.

In a simple leisure app, the difference in speed of recognition may be too subtle to have any noticeable effect on the overall experience.

This may be different for a complex trading application: The requirements for iconography here are more likely to prioritize speed, since every second spent on processing individual elements can have a significant impact on the effectiveness of the overall interface.

Semiotic theory and structure of signs

User interface, like any language or other communication system, is a construct, made using series of signs. Semiotics, a branch in linguistics that studies signs, defines a sign as being composed of two elements–a signifier (the form which the sign takes) and a signified (the concept it represents).1 A sign is a recognizable combination of a signifier with a particular signified.

The same signifier (form) could stand for a number of different signified (concepts), and multiple signifiers could be used to represent the same concept.2 It is this unique pairing that forms a sign and makes it meaningful in a particular context.

Example of signifier (a traffice sign) vs signified (a photo of an at-grade railroad crossing).
Rail crossing signs

For example, the concept of a railway crossing can be represented in a number of different ways. At the same time, numerous interpretations can be suggested for the railway crossing road sign, by someone unfamiliar with the road sign system.

Based on this relationship between a signifier and signified, and their level of convention (symbolism), signs are classified into three categories:

  • iconic: relate to the concept through resemblance
  • symbolic: relate to the concept through convention, and
  • indexical: relate to the object through causation3

Typically UI icons are symbolic or iconic in nature, and many fall somewhere in between.4

Symbolic signs are often arbitrary and represent a concept in somewhat abstract way (using An icon with a circular arrow. to represent ‘refresh’ function). Their meaning is usually difficult to guess and has to be learned.

Iconic signs, on the other hand, can represent concepts in a more literal way and their meaning can often be guessed using common knowledge (such as An icon with a telephone handset. to represent ‘call’ functionality). Both types can adopt metaphors to enhance understanding.

Iconic versus symbolic

Some time ago I carried out a small icon intuitiveness test on Adobe InDesign and Quark XPress toolbars.5 The test participants, who were using the applications for the first time, were asked to guess the possible meaning of the icons.

I observed an interesting (although hardly surprising) pattern: When people were presented with unfamiliar signs, they usually tried to interpret them as iconic–based on the objects they resemble–and use existing common knowledge to understand them.

For example, the iconic ‘button’ sign An icon with an oval button. in the InDesign toolbar was quickly understood and interpreted by all participants correctly. On the other hand, the ‘Free transform tool’ An icon with a dotted box and an arrowhead. was less successful in terms of its communicative function. The latter is a symbolic sign with indexical features.

For designers, it is obvious that the An icon with a dotted box and an arrowhead. icon is based on the graphical representation of the state that indicates an object has been selected for transformation. However, the people new to the jargon of the application tried to interpret it based on the elements of the sign they were already familiar with. Suggested interpretations included ‘cut rectangular shapes,’ ‘resize,’ ‘select an object by choosing points,’ and ‘select points.’

Naturally, interpreting more abstract (symbolic) icons took significantly longer than iconic signs.

Visual style and speed of recognition

If the icons have the same general shape and denote the same concept, what effect does their visual appearance have on the process of interpretation?

The author of an article recently published on Medium claims that a particular style–line icons–are harder for the brain to process than their solid style equivalents, and can ultimately lead to “cognitive fatigue.”

Setting aside cognitive psychology and theories on how brains process shapes, I wanted to compare the two styles of icons using the semiotic approach6 and some basic user testing.

This time, instead of guessing the meaning of icons, I was reading out the object names (e.g. Camera, Mail) and timing how fast people pointed to an icon that represented that concept.

Two versions, line and filled, of the same icon set were compared.

Icon sets shown to participants.
Icon sets shown to participants.

The participants7 were divided into two groups; each group was given one of the icon styles, initially, before trying the other style. The time taken to find and identify the icons is recorded in the table below.

Group A Group B
Filled Line Line Filled
64.2 28.2 72.6 31.2
63.0 27.6 61.8 26.4
71.4 35.46 60.6 27.6

The icons were in the same position for both rounds of the test, so due to the increased familiarity, the speed of recognition naturally increased for the second style, regardless which style it was.

On average, people who were shown the filled set first discovered the icons in 66.2 seconds, and those who were shown the line set first discovered them in 65 seconds. In the second round, the filled set was complete in 28.4 seconds on average, and the line set took about 30.4 seconds. The average increase in speed for the second part of the test was approximately 36 seconds for both groups.

There was no noticeable difference in speed of recognition for the two styles. So we could suggest that, if an icon is well designed and represents a particular recognizable shape clearly, its style doesn’t have a significant effect on the speed of recognition.8 The more notable difference, however, was in how individual signs relate to the concepts they represent–the relation of signifier to signified in particular signs.

The icons that performed faster in both sets were iconic signs–‘calculator,’ ‘camera,’ ‘mail.’ On the other hand, the icons that seemed to have required a short pause were symbolic: ‘download,’ ‘back,’ ‘copy.’ Even the ‘delete’ icon caused a slight delay in participants’ response.

Interestingly, even though some icons were slightly slower to be recognized, they were not always perceived as being harder to find. When interviewed afterwards, participants classified the ‘delete’ icon as being very easy to find even though it took them slightly longer than some of the others. My guess would be that the reason for it is psychological. We experience a certain satisfaction when discovering the signs we learned well, such as a trash can for ‘delete,’ a floppy disk for ‘save,’ and the like. They are like a secret code, and, feeling pleased with ourselves for learning them, we don’t notice that it took slightly longer to recognize.

The table below summarizes the performance of the individual icons in both sets. The icons with a red outline took slightly longer, on average, to recognize in both sets than the ones outlined in green.

The same icon list as prior, with highlighting showing which were harder or easier to find. (Also discussed in text.)
Icons flagged with relative ease/difficulty of identification.

Other factors

Consistency in the use of terminology for the icon labels and functionality they represent also seemed to impact the speed of recognition. For example, those who were asked to find a ‘clock’ completed the task slightly faster than those who were trying to find a ‘recent or history’ function. In both cases, the function was represented by the same ‘clock’ icon.

Another aspect that could influence speed is proximity of similar shapes, as well as proximity of icons which represent related concepts. For example, both flag and folder icons (top right) are based on a similar size rectangular shape, which may have caused a slight delay in response. Another example is ‘important’ (flag) and ‘favorites’ (star)–they represent similar concepts and could therefore be mistaken for one another, particularly if placed close to each other.

Therefore, it was not the style of the icon set but rather the individual icons in both sets, their shapes, and their position in relation to each other that made the most difference in the speed of recognition in this test.

Guidelines for ‘faster’ icons

It is well known that icons with labels are more effective than icons alone, particularly for the learners of the application. This is expected, since a text label is a direct link that serves to bridge the gap between the signifier and signified.

However, sometimes, for whatever reason, you may need to use icons alone or use hover over tooltips only. You may also need those icons to be discovered and recognized quickly. Here are the guidelines when that’s the case.

Two different icons showing the concept of calculator.

When possible, choose iconic signs that relate to concepts through resemblance, rather than symbolic signs. For example, the ‘calculator’ icon on the left is expected to be discovered and recognized faster than the one on the right.9

Icon showing the classic red, octagonal stop sign.

For crucial concepts that must be understood fast and also for abstract concepts, a combination of word labels and color is more effective than an icon. For example, it’s this combination that allows the ‘stop’ road sign to be recognized quickly.

Placing icons that represent similar concepts close to each other–for example, ‘tasks,’ ‘inbox,’ ‘notifications’–may slow down the recognition of each individual icon in the group.

Terminology used in the user interface should relate to the icon labels as closely as possible. For example, if you are using a flag icon to represent ‘important,’ label it ‘flag as important’ rather than ‘mark as important.’

Above all, when testing the icons, the main focus should be not on the style but rather individual shapes of the icons, how they relate to the concepts they represent, and how they work with each other–visually and conceptually.


1This is the original definition by a Swiss linguist Ferdinand de Saussure. Another well known sign structure was suggested by Charles Peirce.

2In fact, the number of combinations is unlimited–which is known in linguistics as unlimited semiosis.

3Although Peirce’s original classification of signs is much more elaborate.

4This classification is made when comparing the interface icons to each other. In a more general sense, all UI icons are essentially iconic signs.

5The test was part of my BA dissertation based on a comparative semiotic analysis of the two interfaces.

6A semiotic approach views an interface as a medium of communication which operates through a form of metalanguage. All elements are seen as encoded signs and all operations represent communicative processes between the designer and the user of the system. The semiotic approach is not an alternative but rather complementary to the user-centred approach. While the choice of signs in the user-centred approach is mainly determined by empirical studies (the sign is considered to be successful when it is recognised by a statistically significant number of users), semiotics aim to provide a theoretical basis to explain the phenomenon of user-centeredness and direct the designer towards the correct choice of signification systems amongst all of the possible signifiers. A book by C. de Souza provides a great insight into the semiotic approach.

7Six people took part in the test (thanks to my brave and adventurous colleagues). While the number of participants was too low to make any definite conclusions, it helped to see the general pattern in interpreting the icons.

8It may be worth mentioning that two of the participants found the full set “easier on the eye.” While this could be the case, it could also be due to the their better familiarity with the solid style. Four other participants did not notice the difference between the two styles, and only mentioned that one was ‘black’ while the other was ‘white’ which they didn’t see as notably different.

9Note: This is based on my assumptions and observations of testing similar icons; those particular examples have not been verified by user testing.


Andersen, P. B. 1990. A Theory of Computer Semiotics. Cambridge: Cambridge University Press.

Barr, P. 2003. User-Interface Metaphors in Theory and Practice

de Souza, C. S., 2004. The Semiotic Engineering of Human-Computer Interaction. Cambridge: The MIT Press.

Eco, U. 1978. A Theory of Semiotics. Bloomington: Indiana University Press.

Peirce’s Theory of Signs

Peirce, Eco, and unlimited semiosis

Peirce, C.S. 1998. The Essential Peirce. Volume 2. Indiana University Press.

Semiotics for beginners

Saussure, F. de. 1983. Course in General Linguistics. London: Duckworth.

Wiedenbeck, S., 1999. The use of icons and labels in an end user application program: an empirical study of learning and retention.


  1. Excellent overview, well-referenced, thanks for links to read more on the topics, and I like that you were able to include numbers from your own research. I think a lot of us who are classically trained graphic designers know most of the basics, but have forgotten the details. I will be referring to this a lot.

    Nitpicks: 1) Stop sign is cropped on the sides. When used as a universally-recognized icon, it sorta messes with it. 2) InDesign is a program, Quark is a manufacturer (Xpress is the program).

  2. Steven Hoober – thanks for pointing out those little things, now corrected.

    Sloan – extremely thoughtful and informative piece by Matt Queen, thank you for referencing.

  3. Alla, loved this article. Would love to know two things in a follow up if you ever get the chance:

    1. Whether the size of the icon impacts recognition speed.
    2. Whether the line-art vs. solid icon recognition results might skew different for different age groups; my specific interest is whether older folks (and those with poor eyesight) would have a tougher time recognizing line art vs. solid art.

    In any case, look forward to reading more from you. Thanks for sharing your knowledge!

    1. UXUiOS – thanks, these are great points to address in the future.

      Last year I did some user testing comparing how people in the age group over 60 use tablets (both Windows RT and iOS). The icon recognition was extremely slow–icons without labels were often overlooked or misinterpreted. Generally people didn’t perceive them as actionable items when trying to accomplish a task, and, in many cases, didn’t notice them.

      Although we can’t make conclusions prior to some thorough research and user testing, my initial guess would be that people who are less experienced in using modern systems may have equal difficulties recognising line and solid icons.

      An excellent article referenced in the previous comment examines in depth the process of image recognition using M and P pathways–where the M pathway “contains channels sensitive to gross shape”, and the P pathway “contains channels sensitive to color and detailed shape.”

      In our case we compare the same monochrome icons, and the only difference is their visual treatment–line or solid. The visual clues (the overall geometric shape and the fine details) that allow them be associated with a particular object or concept should be the same for both styles.

      However, in some icons the line or solid visual treatment may cause those visual clues to be altered–it would be interesting to identify specific examples.

  4. It’s clear that both groups’ average response time was halved in the second trial; this indicates that they were carrying over positional cues from the first trial.

    You may have had significantly different results if the placement (and potentially, layouts) of the icons had been different between the two trials.

  5. Hi, Alla, great article. Can i translate your article to russian and post it on my blog (linking to your site, of course)?

  6. Hi and thanks for this very interristing document.

    I would just want to say it could be interristing to speak about time life of our icon. Because when you show a phone icon in your article almost young people didn ‘t see such a phone because they live with smartphones. In this case we can say we keep this old phone form because it is easier and faster to be identified as a phone rather a smartphone icon….so in reality it becomes a learnt icon atfer being a pictogramme

  7. Great article. I have been wondering recently about the line vs. fill perception as well as in rendering difference in both (another subject entirely), and on perception across ages and cultural context. I have seen just from observation that some of the icons need to be learned and naming the icon with text enables finding it faster and getting used to it. The change in mobile devices operating systems, for example, is another area where icons are constantly changed and relearned. Thank you for posting!

  8. While the understanding and application of semiotic theory here seems a bit crude, it’s an interesting endeavour nonetheless.

    I do find your calculator example a bit contradictory though.
    “It is well known that icons with labels are more effective than icons alone…”
    “When possible, choose iconic signs that relate to concepts through resemblance, rather than symbolic signs.”
    I think one could argue that the mathematical symbols are almost like a label, telling us “math! calculator!” long before a rectangle with other rectangles and squares cut out of it would.
    So if icons with labels are more effective than icons alone it seems contradictory to say that adding symbolic information (like words or mathematical symbols) to an icon is to be avoided.

    1. Ingrid – thanks for reading the article and for sharing your thoughts. My prediction for the calculator icon is based on the assumption that it takes slightly longer to assign a specific (and correct) meaning to an arbitrary symbol than it takes to recognise an object based on resemblance. Reading written words is different to assigning meaning to abstract signs, such as math symbols. The word ‘Calculator’ for example would normally have only one meaning for most people, in the context of a specific interface. On the other hand, the math symbols could create a much larger number of interpretations, e.g. ‘Math functions’, ‘View Calculations’, ‘Trading’, ‘Count’, ‘Measurements’, etc. While the meaning of this icon seems pretty obvious to us, the people who are less familiar with it could suggest different interpretations. The signs that have a potential to generate a number of possible meanings (often symbolic signs) may require additional time to make the connection between the signifier and the right signified, among all the available choices.

      Please note however that this prediction is based on my personal observations and assumptions – as mentioned in the post, this specific icon has not been verified by user testing.

  9. Good article and interesting points. The best of the article is driving at the core differences between symbolic and iconic pictographs to describe an action. This is important data to take into account in the development of an icon set.

    I do have concerns with the method of the testing in your icons sets though:

    1) To Will Parker’s point you’ve created a false association with the speed of the line icons by maintaining the same testing group with associations with shapes already developed. I’d suggest these variables would give a more full understanding of the issues:

    * A wider range of people with varied interface experiences.
    * People with different eyesight capabilities (my own eyes aren’t great for instance)
    * Reviewing inverse color relationships (light on dark, dark on light) for icons.
    * Reviewing line density.
    * Reviewing color grouping (adding oriented actions, subtracting oriented actions, reviewing oriented actions, etc)

    2) To UXUiOS’s point, I’d love to see this test done on not only people of different ages (younger as well), but on people with different eyesight.

    I’m not advocating for one style over the other necessarily. Style is often a choice for the aesthetic of the application in which an icon is being used, whether for wayfinding through an airport or wayfinding through a digital application.

    This is important research. Thanks for getting it going. I’d love to participate or help in the research if you continue.

    1. Matthew Smith – thank you, just to confirm, there were two groups in the test – the first group was shown the filled set first, while the second group saw the line set first – therefore neither of the sets should have had an advantage. Apologies if this hasn’t been made very clear in the article.

      Thank you for the suggestions – they will be extremely helpful when deciding on the direction for the future research. So far my interest has been mainly on the process of assigning meaning to signs, however many other factors, such as the ones you mentioned, may make a significant difference to the conclusions drawn from the study. Identifying and addressing these factors would certainly be required to build a comprehensive analysis.

Comments are closed.