Grouping and visual search: how the eye binds marks into objects, and what that licenses

An interface is a field of marks, and the eye does not see marks. It sees objects. Unit 1 set out what the eye does with the light. This one asks what it does with the marks on the screen. The gestalt laws describe how marks bind into wholes. The visual-search literature describes how a target is found among the distractors. Both tell a designer which way an effect will run, and both are often read as if they fixed its size. They do not, and the strongest counter is that the models have moved past description: Kubovy and van den Berg (2008) measure grouping strength on a ratio scale, and guided search predicts search time from the display (Wolfe, 2021). Treat a layout rule from either as a hypothesis, not a specification.

Marks become objects before you can name them

Wertheimer (1923) showed that the eye organises marks into wholes by rules not present in the marks themselves. He named proximity, similarity and common fate among those rules, and argued from demonstrations rather than measurement. Wagemans and colleagues (2012) reviewed a century of the work. They catalogued the classical principles, among them proximity, similarity, common fate and good continuation, and set newer principles beside them, including synchrony and common region. They report that past experience and attention influence figure-ground organisation, which loosens the picture of a purely bottom-up process.

Work a case in which spacing alone decides the grouping. A settings list spaces every row by the same 12 pixels, and small headings separate the groups. A reader takes the whole list as one block, and the headings stop doing work. Raise the gap between groups to 28 pixels and leave the row gap at 12. The same marks then read as four blocks. The mechanism is proximity. The case shows the direction of the effect, and not its size, because the threshold spacing for this display was never measured.

The laws describe, and a model predicts

Wagemans and colleagues (2012) find that the classical principles describe perceptual organisation rather than explain it, and that competing accounts sit side by side with no single mechanism accepted.

Quantification is possible, and it changes the picture. Kubovy and van den Berg (2008) built a probabilistic model that measures grouping strength on a ratio scale. In three dot-lattice experiments they found that the conjoint effect of proximity and similarity equals the sum of the two separate effects. A quantified model earns the right to predict size. An unquantified principle earns the right to predict direction only.

Grouping is not fixed: memory bends it

Grouping is not fixed from display to display either. Kimchi (1992) reviewed the global-local paradigm, in which a large shape is built from smaller ones. She argued that the advantage of the global level is conditional on the task and the stimuli. The claim that wholes are always seen before their parts does not follow.

Recent work makes the point from the other direction. Koc and colleagues (2025) separated working-memory load from perceptual load, using an N-back task for the first and flanker congruency for the second. They found the orientation bias in grouping by proximity depended on working-memory load and not on perceptual load: what an observer holds in mind changes how the same display groups.

A single feature pops out, and a conjunction is searched

Treisman and Gelade (1980) proposed feature-integration theory, in which single features are registered in parallel while a conjunction of more than one feature must be assembled by focused attention, directed serially from item to item.

Wolfe (1994) revised this into guided search, in which a massively parallel preattentive stage computes basic features and guides a later limited-capacity stage to likely locations. Wolfe and Horowitz (2004) placed colour, motion, orientation and size among the attributes supported by converging evidence, and found that arbitrary shapes and letter identities do not qualify. Colour guides search only where the reader has the channel.

Work a search through. A user looks for a red O among red Ts and green Os. Colour and shape both appear in the display, but no single feature separates the target from the distractors. It is therefore a conjunction target, and search time rises with the number of items, the pattern Treisman and Gelade predicted. Change the distractors to green Ts. The target then becomes the only red mark, it pops out for a reader who separates red from green, and search time barely moves with set size. The pair of displays licenses a target that differs on one guiding attribute. It does not fix how many items a display may carry, which follows the distractors.

The evidence stops short of a specification

Search does not divide cleanly into two modes. Duncan and Humphreys (1989) varied how similar targets were to distractors, and how similar distractors were to each other, and found that search efficiency fell on a continuum set by those similarities rather than by whether a feature or a conjunction is sought.

Work the similarity variable. Hold a single-feature target constant and vary only the distractors. With a few homogeneous distractors the target is found quickly. Make the distractors heterogeneous, so that no two share a feature. Search time then climbs although the target has not changed. Duncan and Humphreys read that as a continuum set by similarity, and not as a strict split between feature and conjunction search. The case shows that cost is a property of the distractors supplied. It does not say which similarity did the work unless target-to-distractor and distractor-to-distractor similarity are varied apart.

Desimone and Duncan (1995) supplied a mechanism consistent with that view. They reviewed the neural evidence and argued that objects compete for limited capacity, and that attention biases that competition rather than passing a beam from item to item. A designer who reads search cost as the price of a serial scan is reading one model as the phenomenon.

The encoding ranking is evidence, not a layout rule

Cleveland and McGill (1984) had participants judge quantities encoded as position, length, angle, area and colour, and ordered the elementary perceptual tasks by accuracy. Position along a common scale is judged most accurately, and colour hue least, and their guideline is to build graphs around tasks high in that ordering. The ranking came from measured error, so it is evidence rather than taste. Work one dataset twice, as bars on a common baseline and as slices of a pie. The measured ordering predicts that a reader recovers a difference from the bars more accurately than from the slices, though it says nothing against a pie that shows shares of a whole. Healey and Enns (2012) argue that display design should rest on measured perceptual and memory limits rather than intuition.

Contested question: is the mechanism fixed, and where does search cost come from?

Two disagreements run through the unit. The first is the shape of search. Treisman and Gelade (1980) hold that features are processed in parallel while conjunctions require serial attention. That view predicts a steep cost for conjunction targets whatever the distractors. Duncan and Humphreys (1989) hold that efficiency is a continuum governed by similarity. Wolfe's guided-search models occupy the middle, and by Wolfe (2021) the account is explicitly a hybrid of serial and parallel processes. The two original camps are evenly matched, each fitting the studies built to test it.

The second disagreement is whether the organising principles are fixed or contextual. Read one way, the gestalt laws are properties of the visual system that apply whenever the stimulus is presented (Wertheimer, 1923). Read another, grouping is conditional on the task and the stimuli (Kimchi, 1992), and working memory modulates it (Koc et al., 2025). Here the contextual reading has the firmer ground, resting on measured interactions rather than demonstrations.

Specific evidence would separate the positions. For search, the test holds the target constant while varying target-distractor similarity and distractor homogeneity independently. It measures response-time distributions and eye movements, because a genuinely serial process leaves a trace that a continuous accumulation of evidence does not. For grouping, the test is a quantified model run as a prediction. State the expected grouping strength for a new display in advance, then check it under varied working-memory load. Until such tests are run, both literatures predict direction, and the size of any effect is open.

While the dispute is open, a designer can act on what both readings allow. Group related items so that the grouping reads without a label. Test a layout against the display it will carry, because search cost follows the distractors and not only the target.

Where the argument is heading

The direction of travel is towards models that predict magnitude, and towards treating a layout rule as a claim with a test attached. Wolfe's (2021) Guided Search 6.0 combines five sources of guidance, and Koc and colleagues (2025) tie perceptual organisation to working memory.

The sources this unit rests on

• Cleveland, W. S. and McGill, R. (1984). Graphical perception: theory, experimentation, and application to the development of graphical methods. Journal of the American Statistical Association, 79(387), 531-554. doi:10.1080/01621459.1984.10478080 Ordered the elementary graphical-perception tasks by accuracy, from position along a common scale at the top to colour hue at the bottom, and advised building graphs around tasks high in that ordering. It is the unit's measured basis for preferring some encodings to others.
• Desimone, R. and Duncan, J. (1995). Neural mechanisms of selective visual attention. Annual Review of Neuroscience, 18, 193-222. doi:10.1146/annurev.ne.18.030195.001205 Reviews the neural evidence for biased competition, in which objects compete for limited capacity and attention biases the competition rather than scanning item by item. It makes a strict serial-scan reading of search hard to sustain.
• Duncan, J. and Humphreys, G. W. (1989). Visual search and stimulus similarity. Psychological Review, 96(3), 433-458. doi:10.1037/0033-295x.96.3.433 Varied target-distractor and distractor-distractor similarity in visual search and found a continuum of efficiency set by those similarities, supporting neither a serial-parallel nor a feature-conjunction distinction. It is the central counterweight to the two-mode view.
• Healey, C. G. and Enns, J. T. (2012). Attention and visual memory in visualization and computer graphics. IEEE Transactions on Visualization and Computer Graphics, 18(7), 1170-1188. doi:10.1109/tvcg.2011.127 Surveys attention and visual-memory research for visualisation and argues that display design should rest on measured perceptual and memory limits, bridging laboratory findings to interface practice.
• Kimchi, R. (1992). Primacy of wholistic processing and global/local paradigm: a critical review. Psychological Bulletin, 112(1), 24-38. doi:10.1037/0033-2909.112.1.24 Reviews the global-local paradigm and argues that the global advantage is conditional, so it is not evidence that wholes are always perceived before their parts. It stops the unit treating the grouping laws as unconditional.
• Koc, A., Bulajic, A., Esposito, A., Lachmann, T. and van Leeuwen, C. (2025). Perceptual bias in Gestalt grouping by proximity affected by visual working memory load. Acta Psychologica, 260, 105102. doi:10.1016/j.actpsy.2025.105102 Separated working-memory load from perceptual load and found the orientation bias in grouping by proximity depended on working-memory load, not perceptual load. So grouping depends on what the observer holds in mind, and is not fixed across contexts.
• Kubovy, M. and van den Berg, M. (2008). The whole is equal to the sum of its parts: a probabilistic model of grouping by proximity and similarity in regular patterns. Psychological Review, 115(1), 131-154. doi:10.1037/0033-295x.115.1.131 Built a probabilistic model measuring grouping strength on a ratio scale and found the conjoint effect of proximity and similarity equals the sum of their separate effects. It shows a grouping law can predict magnitude, not only direction, once stated quantitatively.
• Treisman, A. M. and Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology, 12(1), 97-136. doi:10.1016/0010-0285(80)90005-5 Proposed feature-integration theory, in which features are registered in parallel while conjunctions need focused serial attention, and tested the predictions across search, texture segregation and localisation. It is the founding two-mode account behind the unit's pop-out rules.
• Wagemans, J., Elder, J. H., Kubovy, M., Palmer, S. E., Peterson, M. A., Singh, M. and von der Heydt, R. (2012). A century of Gestalt psychology in visual perception I: perceptual grouping and figure-ground organisation. Psychological Bulletin, 138(6), 1172-1217. doi:10.1037/a0029333 Reviews a century of grouping and figure-ground work, cataloguing the classical principles and newer ones such as synchrony and common region, and reporting where the evidence is strong or thin. It is the unit's authority for what the laws are.
• Wagemans, J., Feldman, J., Gepshtein, S., Kimchi, R., Pomerantz, J. R., van der Helm, P. A. and van Leeuwen, C. (2012). A century of Gestalt psychology in visual perception II: conceptual and theoretical foundations. Psychological Bulletin, 138(6), 1218-1252. doi:10.1037/a0029334 Reviews the conceptual foundations and the competing computational accounts of grouping, including the principle that the brain favours the simplest available organisation. It shows the principles describe organisation while their explanation is contested, so they name an effect rather than fix its size.
• Wertheimer, M. (1923). Untersuchungen zur Lehre von der Gestalt II. Psychologische Forschung, 4(1), 301-350. doi:10.1007/bf00410640 Described grouping by proximity, similarity and common fate, showing the eye organises marks by rules not present in the marks themselves. It argues from demonstrations, so it establishes that grouping happens while leaving its size unstated.
• Wolfe, J. M. (1994). Guided Search 2.0: a revised model of visual search. Psychonomic Bulletin and Review, 1(2), 202-238. doi:10.3758/bf03200774 Presents guided search, in which a parallel preattentive stage computes basic features and guides a limited-capacity stage to likely locations, replacing a strict parallel-serial division. It is the working model the unit teaches.
• Wolfe, J. M. and Horowitz, T. S. (2004). What attributes guide the deployment of visual attention and how do they do it? Nature Reviews Neuroscience, 5(6), 495-501. doi:10.1038/nrn1411 Sorts stimulus attributes by the evidence that they guide attention, placing colour, motion, orientation and size among those supported while rejecting arbitrary shapes and letter identities. Search efficiency, the slope of response time against set size, is the criterion.
• Wolfe, J. M. (2021). Guided Search 6.0: an updated model of visual search. Psychonomic Bulletin and Review, 28(4), 1060-1092. doi:10.3758/s13423-020-01859-9 Updates guided search to five sources of guidance combined into a priority map, sends attention to the most active location about twenty times a second, and models recognising an attended item as a diffusion process taking more than 150 milliseconds. The account is now explicitly a hybrid of serial and parallel processes.